CN103106265A - Method and system of classifying similar images - Google Patents
Method and system of classifying similar images Download PDFInfo
- Publication number
- CN103106265A CN103106265A CN2013100377412A CN201310037741A CN103106265A CN 103106265 A CN103106265 A CN 103106265A CN 2013100377412 A CN2013100377412 A CN 2013100377412A CN 201310037741 A CN201310037741 A CN 201310037741A CN 103106265 A CN103106265 A CN 103106265A
- Authority
- CN
- China
- Prior art keywords
- image
- identified
- response diagram
- response
- gradient
- 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
Images
Landscapes
- Image Analysis (AREA)
Abstract
The invention provides a method and a system of classifying similar images. The method comprises the following steps: inputting images to be identified and obtaining shape features, gradient features, color features and texture features of the images to be identified; segmenting a training sample in an image base into a plurality of local area images different in size, and conducting size conversion so as to obtain an image template set which comprises a plurality of image templates; analyzing the image templates in the image template set and obtaining shape features, gradient features, color features and texture features of the image templates; matching and processing corresponding features of the images to be identified and the images in the image template set so as to obtain image detailed information of the images to be identified; obtaining classifications of the images to be identified through image presenting data and by utilization of a bagging classifier. Accurate classification of similar images is realized through extracting and matching the shape features, the gradient features, the color features and the texture features of the input images according to the method.
Description
Technical field
The present invention relates to the image recognition technology field, particularly a kind of similar image sorting technique and system.
Background technology
Classification is a kind of cascade form of tree structure in essence, in this structure, describes near the high one-level node of root and merges classification (inclusive class), and also referred to as upper classification (super-ordinate class), for example, vehicle etc.The middle layer also makes basic class (basic class) node describe more concrete classification, for example, and motorcycle or automobile etc.And near the low node layer of leaf node, also referred to as subordinate classification (subordinate class), usually catch between object fine distinction more, and for example, motion motorcycle or multifunctional motorcycle, passenger vehicle, truck or car etc.Similar image classification refers to be classified in same basic class or shape and the aspect such as vision is apparent and close object thereof, namely the object of subordinate classification is classified, and for example, distinguishes different types of mushroom, automobile etc.
Along with the develop rapidly of computer technology, artificial intelligence technology and sensor technology, and the mankind are in the active demand in work, studying and living field, and the pattern classification technology has developed into a brand-new stage.But at present expert and scholar concentrate on focus in classification work to basic layer (basic level) object mostly, and seldom mention similar image, i.e. the classification of subordinate layer (subordinate level) object.
Traditional method for classifying modes is applied to often meeting failure in the high similar image classification of similarity, and its main cause has following several:
The first, some classical sorting techniques are often extracted feature with codebook mode at present, and this " dictionary " builds with non-supervisory method usually." entry " that use k average or gauss hybrid models cluster obtain adjudicated not necessarily Useful Information to classification in fact although sometimes occur at certain area of space high probability.The second, when detected zone was mapped to " dictionary entry " form, a lot of detailed information were lost.The 3rd, " dictionary " method needs some parameters of manual adjustments cluster, and is both loaded down with trivial details, not necessarily selects again suitablely especially.And in contrast to this, made up to a great extent the deficiency of above-mentioned use code book method based on the method for annotations, and recognition effect is also very good, but huge cost of labor is limited by very large its development.
Effectively solve basic layer classification problem without the Supervisory Shell linked method, yet can not distinguish the very large subordinate layer classification of correlativity.Recognition methods based on attribute has also shown very large advantage.These technology are normally utilized the training data study identification model with attribute tags, the then appropriate perceptual property of Applied Learning model evaluation test pattern.Like fur, point or four attributes such as leg are effectively to these methods for recognition category, but between the subordinate layer object, fine distinction but seems not satisfactory for distinguishing.
Summary of the invention
Purpose of the present invention is intended to solve at least one of above-mentioned technological deficiency.
For achieving the above object, the embodiment of one aspect of the present invention proposes a kind of similar image sorting technique, comprises the following steps: S1: input image to be identified and obtain shape facility, Gradient Features, color characteristic and the textural characteristics of described image to be identified; S2: the training sample in image library is cut apart the regional area image that generates a plurality of different sizes, and carried out change of scale, obtain the image template collection, described image template is concentrated and is comprised a plurality of image templates; S3: analyze the image template that described image template is concentrated, and obtain shape facility, Gradient Features, color characteristic and the textural characteristics of image template; S4: the character pair of described image to be identified and described image template collection image is mated, and process, obtain the image detail information of described image to be identified; S5: by described image representation data and utilize the Bagging sorter to obtain the classification of image to be identified.
Method according to the embodiment of the present invention, by input picture being carried out extraction and the coupling of shape facility, Gradient Features, color characteristic and textural characteristics, the factors such as affine, illumination have effectively been overcome to the image of classification results, guarantee simultaneously image representation information integrity, rich and discriminability, guaranteed the correct classification of similar image.
In an example of the present invention, described step S2 specifically comprises: S21: the training sample in image library is cut apart the regional area image that generates a plurality of different sizes, consisted of the first image template collection; S22: each image template that described the first image template is concentrated carries out change of scale, obtains the image template of different scale, the composing images template set.
In an example of the present invention, described step S4 specifically comprises: S41: concentrate the character pair of image to mate described image to be identified and described image template, obtain Characteristic of Image response diagram to be identified, wherein, described Characteristic of Image response diagram to be identified comprises shape response diagram, gradient response diagram, color response figure and texture response diagram; S42: with the numerical value in every width response diagram of shape response diagram, gradient response diagram, color response figure and the texture response diagram of described image to be identified by sequence from big to small, then get front a plurality of data in sorted numerical value, form the First Characteristic collection of this width response diagram; S43: every width response diagram of shape response diagram, gradient response diagram, color response figure and the texture response diagram of described image to be identified is divided into respectively a plurality of zones, and the numerical value in each zone in described every width response diagram is pressed from big to small sorted respectively, then a plurality of data before getting respectively in sorted numerical value, the data of in certain sequence All Ranges being taken out form the Second Characteristic collection of this width response diagram; S44: First Characteristic collection and the Second Characteristic collection of described every width response diagram are connected, the composition characteristic vector, with the proper vector series connection of all shape response diagrams, the proper vector series connection of all gradient response diagrams, the proper vector series connection of all colours response diagram and the proper vector series connection of all texture response diagrams, obtain shape blending proper vector, gradient fusion feature vector, color blend proper vector and the texture fusion feature vector of image to be identified respectively; S45: shape blending proper vector, gradient fusion feature vector, color blend proper vector and the texture fusion feature vector of described image to be identified are merged again, 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 comprises: S51: the objective function of design Bagging sorter; S52: by described objective function and utilize described image detail information to obtain the proper vector weight sets; S53: according to a plurality of sorters, the degree of confidence of described proper vector weight sets is determined classification and the output of described image to be identified.
For achieving the above object, embodiments of the invention propose a kind of similar image categorizing system on the other hand, comprising: load module is used for shape facility, Gradient Features, color characteristic and the textural characteristics inputting image to be identified and obtain described image to be identified; Divide module, be used for the training sample of image library is cut apart the regional area image that generates a plurality of different sizes, and the regional area image is carried out respectively change of scale, obtain the image template collection, described image template is concentrated and is comprised a plurality of image templates; Analysis module is used for analyzing the image template that described image template is concentrated, and obtains shape facility, Gradient Features, color characteristic and the textural characteristics of image template; Matching module is used for concentrating the character pair of image to mate described image to be identified and described image template, and processes, and obtains the image detail information of described image to be identified; Obtain module, be used for by described image representation data and utilize the Bagging sorter to obtain the classification of image to be identified.
Method according to the embodiment of the present invention, by input picture being carried out extraction and the coupling of shape facility, Gradient Features, color characteristic and textural characteristics, the factors such as affine, illumination have effectively been overcome to the image of classification results, guarantee simultaneously image representation information integrity, rich and discriminability, guaranteed the correct classification of similar image.
In an example of the present invention, described division module comprises: extraction unit, be used for the training sample of image library is cut apart the regional area image that generates a plurality of different sizes, and consist of the first image template collection; The change of scale unit is used for each image template that described the first image template is concentrated is carried out change of scale, obtains the image template of different scale, the composing images template set.
In an example of the present invention, described matching module comprises: matching unit, be used for the character pair of described image to be identified and described image template collection image is mated, obtain respectively Characteristic of Image response diagram to be identified, wherein, described Characteristic of Image response diagram to be identified comprises shape response diagram, gradient response diagram, color response figure and texture response diagram; Search the unit, be used for the numerical value of every width response diagram of shape response diagram, gradient response diagram, color response figure and texture response diagram that will described image to be identified by sorting from big to small, then get front a plurality of data in sorted numerical value, form the First Characteristic collection of this width response diagram; Division unit, be used for every width response diagram of shape response diagram, gradient response diagram, color response figure and the texture response diagram of described image to be identified is divided into respectively a plurality of zones, and the numerical value in each zone in described every width response diagram is pressed from big to small sorted respectively, then a plurality of data before getting respectively in sorted numerical value, the data of in certain sequence All Ranges being taken out form the Second Characteristic collection of this width response diagram; Integrated unit is used for described all features are carried out effective integration, obtains final fusion feature vector, is used for representing image detail information to be identified; The first integrated unit, be used for First Characteristic collection and the Second Characteristic collection of described every width response diagram are connected, the generating feature vector, with the proper vector series connection of all shape response diagrams, the proper vector series connection of all gradient response diagrams, the proper vector series connection of all colours response diagram and the proper vector series connection of all texture response diagrams, obtain shape blending proper vector, gradient fusion feature vector, color blend proper vector and the texture fusion feature vector of image to be identified respectively; And second integrated unit, be used for shape blending proper vector, gradient fusion feature vector, color blend proper vector and the texture fusion feature vector of described image to be identified are merged again, generate final fusion feature vector, be used for representing image detail information to be identified.
In an example of the present invention, described acquisition module comprises: design cell, for the objective function of design Bagging sorter; Obtain the unit, be used for by described objective function and utilize described image detail information to obtain the proper vector weight sets; Determine output unit, be used for according to a plurality of sorters, the degree of confidence of described proper vector weight sets being determined classification and the output of described image to be identified.
The aspect that the present invention adds and advantage part in the following description provide, and part will become obviously from the following description, or recognize by practice of the present invention.
Description of drawings
Above-mentioned and/or the additional aspect of the present invention and advantage will become from the following description of the accompanying drawings of embodiments and obviously and easily understand, wherein:
Fig. 1 is the process flow diagram of similar image sorting technique according to an embodiment of the invention;
Fig. 2 is the comparison diagram of difference between difference and class in class according to an embodiment of the invention;
Fig. 3 is the frame diagram of similar image categorizing system according to an embodiment of the invention;
Fig. 4 is the structured flowchart of matching module according to an embodiment of the invention.
Embodiment
The below describes embodiments of the invention in detail, and the example of embodiment is shown in the drawings, and wherein same or similar label represents same or similar element or the element with identical or similar functions from start to finish.Be exemplary below by the embodiment that is described with reference to the drawings, only be used for explaining the present invention, and can not be interpreted as limitation of the present invention.
Fig. 1 is the process flow diagram of similar image sorting technique according to an embodiment of the invention.As shown in Figure 1, the similar image sorting technique according to the embodiment of the present invention comprises the following steps:
Step S101 inputs image to be identified and obtains shape facility, Gradient Features, color characteristic and the textural characteristics of image to be identified.
Particularly, because the SIFT operator changes, blocks with noise etc. and have good unchangeability translation, rotation, yardstick convergent-divergent, brightness, visible change, affined transformation are also kept to a certain degree stability, so the present invention extracts the SIFT operator as the picture shape feature.LBP has rotational invariance, can overcome to a great extent illumination variation to the impact of image simultaneously, therefore extracts the LBP operator and represents as image texture characteristic.In addition, pertinent literature shows, the gradient of image and color are that image information is the most effectively described, so Simultaneous Extracting Image Gradient Features and color characteristic presentation video.
Step S102 is cut apart the training sample in image library the regional area image that generates a plurality of different sizes, and is carried out change of scale, obtains the image template collection, and image template is concentrated and comprised a plurality of image templates.
Particularly, training sample in image library is cut apart the regional area image that generates a plurality of different sizes, consisted of the first image template collection, and each image template that the first image template is concentrated carries out change of scale, obtain the image template of different scale, the composing images template set.Wherein, the size dimension of template image can be adjusted and standard according to actual needs.
Step S103, the image template in the analysis image template set, shape facility, Gradient Features, color characteristic and the textural characteristics of acquisition image template.
Particularly, because the SIFT operator changes, blocks with noise etc. and have good unchangeability translation, rotation, yardstick convergent-divergent, brightness, visible change, affined transformation are also kept to a certain degree stability, so the present invention extracts the SIFT operator as the picture shape feature.LBP has rotational invariance, can overcome to a great extent illumination variation to the impact of image simultaneously, therefore extracts the LBP operator and represents as image texture characteristic.In addition, pertinent literature shows, the gradient of image and color are that image information is the most effectively described, so Simultaneous Extracting Image Gradient Features and color characteristic presentation video.
Step S104 concentrates the character pair of image mate and process image to be identified and image template, obtains the image detail information of image to be identified.
Particularly, concentrate the character pair of image to mate image to be identified and image template, obtain respectively shape response diagram, gradient response diagram, color response figure and the texture response diagram of image to be identified.then, shape response diagram with image to be identified, the gradient response diagram, numerical value in every width response diagram of color response figure and texture response diagram is by sequence from big to small, then a plurality of data before getting in sorted numerical value, the First Characteristic collection that forms this width response diagram, and the shape response diagram with image to be identified, the gradient response diagram, every width response diagram of color response figure and texture response diagram is divided into respectively a plurality of zones, and the numerical value in each zone in every width response diagram is pressed from big to small sorted respectively, then a plurality of data before getting respectively in sorted numerical value, the data of in certain sequence All Ranges being taken out form the Second Characteristic collection of this width response diagram.Afterwards First Characteristic collection and the Second Characteristic collection of every width response diagram are connected, the composition characteristic vector, with the proper vector series connection of all shape response diagrams, the proper vector series connection of all gradient response diagrams, the proper vector series connection of all colours response diagram and the proper vector series connection of all texture response diagrams, obtain shape blending proper vector, gradient fusion feature vector, color blend proper vector and the texture fusion feature vector of image to be identified respectively.At last, the shape blending proper vector of image to be identified, gradient fusion feature vector, color blend proper vector and texture fusion feature vector are merged again, generates final fusion feature vectorial, be used for representing image detail information to be identified.
Step S105 is by the image representation data and utilize the Bagging sorter to obtain the classification of image to be identified.
Particularly, the objective function of design Bagging sorter, its objective function is as follows:
Wherein, W=(w
1, w
2..., w
p) be the feature set weight, U=[u
pq]=W
TW, u
pqBe the element of matrix U, i.e. W
TThe element of W, l () expression cost loss function,
Expression training plan image set, N is the training sample number, x
iThe proper vector that is i width training image represents, y
iBe the class label, P is the sorter number that the Bagging algorithm comprises, and α and β are regularization parameter, respectively heavy sparse property and the orthogonality of the controlling feature centralization of state power.
Then by objective function and utilize image detail information to obtain the proper vector weight sets, and the degree of confidence of proper vector weight sets is determined classification and the output of image to be identified according to a plurality of sorters.
Fig. 2 is the comparison diagram of difference between difference and class in class according to an embodiment of the invention.As shown in Figure 2, in figure (a) and (c) be dandelion, (b) be coltsfoot.After being classified, similar image can determine accurately the specific category of respective image.
Method according to the embodiment of the present invention, by input picture being carried out extraction and the coupling of shape facility, Gradient Features, color characteristic and textural characteristics, the factors such as affine, illumination have effectively been overcome to the image of classification results, guarantee simultaneously image representation information integrity, rich and discriminability, guaranteed the correct classification of similar image.
Fig. 3 is the frame diagram of similar image categorizing system in accordance with another embodiment of the present invention.As shown in Figure 3, the similar image categorizing system according to the embodiment of the present invention comprises load module 100, division module 200, analysis module 300, matching module 400 and obtains module 500.
Load module 100 is used for shape facility, Gradient Features, color characteristic and the textural characteristics inputting image to be identified and obtain image to be identified.
In one embodiment of the invention, because the SIFT operator changes, blocks with noise etc. and have good unchangeability translation, rotation, yardstick convergent-divergent, brightness, visible change, affined transformation are also kept to a certain degree stability, so the present invention extracts the SIFT operator as the picture shape feature.LBP has rotational invariance, can overcome to a great extent illumination variation to the impact of image simultaneously, therefore extracts the LBP operator and represents as image texture characteristic.In addition, pertinent literature shows, the gradient of image and color are that image information is the most effectively described, so Simultaneous Extracting Image Gradient Features and color characteristic presentation video.
Divide module 200 and be used for the training sample of image library is cut apart the regional area image that generates a plurality of different sizes, and the regional area image is carried out respectively change of scale, obtain the image template collection, image template is concentrated and is comprised a plurality of image templates.
In one embodiment of the invention, divide module 200 and comprise extraction unit 210 and change of scale unit 220.
Extraction unit 210 is used for the training sample of image library is cut apart the regional area image that generates a plurality of different sizes, consists of the first image template collection.
Change of scale unit 220 is used for each image template that the first image template is concentrated is carried out change of scale, obtains the image template of different scale, the composing images template set.
In one embodiment of the invention, training sample in image library is cut apart the regional area image that generates a plurality of different sizes, consist of the first image template collection, and each image template that the first image template is concentrated carries out change of scale, obtain the image template of different scale, the composing images template set.Wherein, the size dimension of template image can be adjusted and standard according to actual needs.
Analysis module 300 is used for the image template of analysis image template set, and obtains shape facility, Gradient Features, color characteristic and the textural characteristics of image template.
In one embodiment of the invention, because the SIFT operator changes, blocks with noise etc. and have good unchangeability translation, rotation, yardstick convergent-divergent, brightness, visible change, affined transformation are also kept to a certain degree stability, so the present invention extracts the SIFT operator as the picture shape feature.LBP has rotational invariance, can overcome to a great extent illumination variation to the impact of image simultaneously, therefore extracts the LBP operator and represents as image texture characteristic.In addition, pertinent literature shows, the gradient of image and color are that image information is the most effectively described, so Simultaneous Extracting Image Gradient Features and color characteristic presentation video.
Matching module 400 is used for concentrating the character pair of image to mate image to be identified and image template, and processes, and obtains the image detail information of image to be identified.
In one embodiment of the invention, matching module 400 comprise matching unit 410, search unit 420, division unit 430, the first integrated unit 440 and the second integrated unit 450.
Matching unit 410 is used for the character pair of image to be identified and image template collection image is mated, obtain respectively Characteristic of Image response diagram to be identified, wherein, Characteristic of Image response diagram to be identified comprises shape response diagram, gradient response diagram, color response figure and texture response diagram.
Search unit 420 and be used for numerical value with every width response diagram of shape response diagram, gradient response diagram, color response figure and the texture response diagram of image to be identified by sequence from big to small, then get front a plurality of data in sorted numerical value, form the First Characteristic collection of this width response diagram.
Division unit 430 is used for every width response diagram of shape response diagram, gradient response diagram, color response figure and the texture response diagram of image to be identified is divided into respectively a plurality of zones, and the numerical value in each zone in every width response diagram is pressed from big to small sorted respectively, then a plurality of data before getting respectively in sorted numerical value, the data of in certain sequence All Ranges being taken out form the Second Characteristic collection of this width response diagram.
The first integrated unit 440 is used for First Characteristic collection and the Second Characteristic collection of every width response diagram are connected, the generating feature vector, with the proper vector series connection of all shape response diagrams, the proper vector series connection of all gradient response diagrams, the proper vector series connection of all colours response diagram and the proper vector series connection of all texture response diagrams, obtain shape blending proper vector, gradient fusion feature vector, color blend proper vector and the texture fusion feature vector of image to be identified respectively.
The second integrated unit 450 is used for the shape blending proper vector of image to be identified, gradient fusion feature vector, color blend proper vector and texture fusion feature vector are merged again, generate final fusion feature vector, be used for representing image detail information to be identified.
In one embodiment of the invention, concentrate the character pair of image to mate image to be identified and image template, obtain respectively shape response diagram, gradient response diagram, color response figure and the texture response diagram of image to be identified.then, shape response diagram with image to be identified, the gradient response diagram, numerical value in every width response diagram of color response figure and texture response diagram is by sequence from big to small, then a plurality of data before getting in sorted numerical value, the First Characteristic collection that forms this width response diagram, and the shape response diagram with image to be identified, the gradient response diagram, every width response diagram of color response figure and texture response diagram is divided into respectively a plurality of zones, and the numerical value in each zone in every width response diagram is pressed from big to small sorted respectively, then a plurality of data before getting respectively in sorted numerical value, the data of in certain sequence All Ranges being taken out form the Second Characteristic collection of this width response diagram.Afterwards First Characteristic collection and the Second Characteristic collection of every width response diagram are connected, the composition characteristic vector, with the proper vector series connection of all shape response diagrams, the proper vector series connection of all gradient response diagrams, the proper vector series connection of all colours response diagram and the proper vector series connection of all texture response diagrams, obtain shape blending proper vector, gradient fusion feature vector, color blend proper vector and the texture fusion feature vector of image to be identified respectively.At last, the shape blending proper vector of image to be identified, gradient fusion feature vector, color blend proper vector and texture fusion feature vector are merged again, generates final fusion feature vectorial, be used for representing image detail information to be identified.
Obtaining module 500 is used for by the image representation data and utilizes the Bagging sorter to obtain the classification of image to be identified.
In one embodiment of the invention, obtaining module 500 comprises design cell 510, obtains unit 520 and definite output unit 530.
Design cell 510 is used for the objective function of design Bagging sorter.Its objective function is as follows:
Wherein, W=(w
1, w
2..., w
p) be the feature set weight, U=[u
pq]=W
TW, u
pqBe the element of matrix U, i.e. W
TThe element of W, l () expression cost loss function,
Expression training plan image set, N is the training sample number, x
iBe the vector representation of i width Characteristic of Image, y
iBe the class label, P is the sorter number that the Bagging algorithm comprises, and α and β are regularization parameter, respectively heavy sparse property and the orthogonality of the controlling feature centralization of state power.
Obtaining unit 520 is used for by objective function and utilizes image detail information to obtain the proper vector weight sets.
Determine that output unit 530 is used for according to a plurality of sorters, the degree of confidence of proper vector weight sets being determined classification and the output of image to be identified.
Fig. 2 is the comparison diagram of difference between difference and class in class according to an embodiment of the invention.As shown in Figure 2, in figure (a) and (c) be dandelion, (b) be coltsfoot.After being classified, similar image can determine accurately the specific category of respective image.
Method according to the embodiment of the present invention, by input picture being carried out extraction and the coupling of shape facility, Gradient Features, color characteristic and textural characteristics, the factors such as affine, illumination have effectively been overcome to the image of classification results, guarantee simultaneously image representation information integrity, rich and discriminability, guaranteed the correct classification of similar image.
Although the above has illustrated and has described embodiments of the invention, be understandable that, above-described embodiment is exemplary, can not be interpreted as limitation of the present invention, those of ordinary skill in the art is not in the situation that break away from principle of the present invention and aim can change above-described embodiment within the scope of the invention, modification, replacement and modification.
Claims (8)
1. a similar image sorting technique, is characterized in that, comprises the following steps:
S1: input image to be identified and obtain shape facility, Gradient Features, color characteristic and the textural characteristics of described image to be identified;
S2: the training sample in image library is cut apart the regional area image that generates a plurality of different sizes, and carried out change of scale, obtain the image template collection, described image template is concentrated and is comprised a plurality of image templates;
S3: analyze the image template that described image template is concentrated, and obtain shape facility, Gradient Features, color characteristic and the textural characteristics of image template;
S4: the character pair of described image to be identified and described image template collection image is mated, and process, obtain the image detail information of described image to be identified;
S5: by described image representation data and utilize the Bagging sorter to obtain the classification of image to be identified.
2. similar image sorting technique according to claim 1, is characterized in that, described step S2 specifically comprises:
S21: the training sample in image library is cut apart the regional area image that generates a plurality of different sizes, consisted of the first image template collection;
S22: each image template that described the first image template is concentrated carries out change of scale, obtains the image template of different scale, the composing images template set.
3. similar image sorting technique according to claim 1, is characterized in that, described step S4 specifically comprises:
S41: concentrate the character pair of image to mate described image to be identified and described image template, obtain Characteristic of Image response diagram to be identified, wherein, described Characteristic of Image response diagram to be identified comprises shape response diagram, gradient response diagram, color response figure and texture response diagram;
S42: with the numerical value in every width response diagram of shape response diagram, gradient response diagram, color response figure and the texture response diagram of described image to be identified by sequence from big to small, then get front a plurality of data in sorted numerical value, form the First Characteristic collection of this width response diagram;
S43: every width response diagram of shape response diagram, gradient response diagram, color response figure and the texture response diagram of described image to be identified is divided into respectively a plurality of zones, and the numerical value in each zone in described every width response diagram is pressed from big to small sorted respectively, then a plurality of data before getting respectively in sorted numerical value, the data of in certain sequence All Ranges being taken out form the Second Characteristic collection of this width response diagram;
S44: First Characteristic collection and the Second Characteristic collection of described every width response diagram are connected, the composition characteristic vector, with the proper vector series connection of all shape response diagrams, the proper vector series connection of all gradient response diagrams, the proper vector series connection of all colours response diagram and the proper vector series connection of all texture response diagrams, obtain shape blending proper vector, gradient fusion feature vector, color blend proper vector and the texture fusion feature vector of image to be identified respectively;
S45: shape blending proper vector, gradient fusion feature vector, color blend proper vector and the texture fusion feature vector of described image to be identified are merged again, generate final fusion feature vector, be used for representing image detail information to be identified.
4. similar image sorting technique according to claim 1, is characterized in that, described step S5 specifically comprises:
S51: the objective function of design Bagging sorter;
S52: by described objective function and utilize described image detail information to obtain the proper vector weight sets;
S53: according to a plurality of sorters, the degree of confidence of described proper vector weight sets is determined classification and the output of described image to be identified.
5. a similar image categorizing system, is characterized in that, comprising:
Load module is used for shape facility, Gradient Features, color characteristic and the textural characteristics inputting image to be identified and obtain described image to be identified;
Divide module, be used for the training sample of image library is cut apart the regional area image that generates a plurality of different sizes, and the regional area image is carried out respectively change of scale, obtain the image template collection, described image template is concentrated and is comprised a plurality of image templates;
Analysis module is used for analyzing the image template that described image template is concentrated, and obtains shape facility, Gradient Features, color characteristic and the textural characteristics of image template;
Matching module is used for concentrating the character pair of image to mate described image to be identified and described image template, and processes, and obtains the image detail information of described image to be identified;
Obtain module, be used for by described image representation data and utilize the Bagging sorter to obtain the classification of image to be identified.
6. similar image categorizing system according to claim 5, is characterized in that, described division module comprises:
Extraction unit is used for the training sample of image library is cut apart the regional area image that generates a plurality of different sizes, consists of the first image template collection;
The change of scale unit is used for each image template that described the first image template is concentrated is carried out change of scale, obtains the image template of different scale, the composing images template set.
7. similar image categorizing system according to claim 5, is characterized in that, described matching module comprises:
Matching unit, be used for the character pair of described image to be identified and described image template collection image is mated, obtain respectively Characteristic of Image response diagram to be identified, wherein, described Characteristic of Image response diagram to be identified comprises shape response diagram, gradient response diagram, color response figure and texture response diagram;
Search the unit, be used for the numerical value of every width response diagram of shape response diagram, gradient response diagram, color response figure and texture response diagram that will described image to be identified by sorting from big to small, then get front a plurality of data in sorted numerical value, form the First Characteristic collection of this width response diagram;
Division unit, be used for every width response diagram of shape response diagram, gradient response diagram, color response figure and the texture response diagram of described image to be identified is divided into respectively a plurality of zones, and the numerical value in each zone in described every width response diagram is pressed from big to small sorted respectively, then a plurality of data before getting respectively in sorted numerical value, the data of in certain sequence All Ranges being taken out form the Second Characteristic collection of this width response diagram;
The first integrated unit, be used for First Characteristic collection and the Second Characteristic collection of described every width response diagram are connected, the generating feature vector, with the proper vector series connection of all shape response diagrams, the proper vector series connection of all gradient response diagrams, the proper vector series connection of all colours response diagram and the proper vector series connection of all texture response diagrams, obtain shape blending proper vector, gradient fusion feature vector, color blend proper vector and the texture fusion feature vector of image to be identified respectively; And
The second integrated unit, be used for shape blending proper vector, gradient fusion feature vector, color blend proper vector and the texture fusion feature vector of described image to be identified are merged again, generate final fusion feature vector, be used for representing image detail information to be identified.
8. similar image categorizing system according to claim 5, is characterized in that, described acquisition module comprises:
Design cell is for the objective function of design Bagging sorter;
Obtain the unit, be used for by described objective function and utilize described image detail information to obtain the proper vector weight sets;
Determine output unit, be used for according to a plurality of sorters, the degree of confidence of described proper vector weight sets being determined classification and the output of described image to be identified.
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 true CN103106265A (en) | 2013-05-15 |
| CN103106265B 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) |
Cited By (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103530645A (en) * | 2013-09-25 | 2014-01-22 | 北京工商大学 | Texture image classifying method and system based on local binary pattern and Zernike moment |
| CN104217221A (en) * | 2014-08-27 | 2014-12-17 | 重庆大学 | Method for detecting calligraphy and paintings based on textural features |
| 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 |
| CN105809182A (en) * | 2014-12-31 | 2016-07-27 | 中国科学院深圳先进技术研究院 | Image classification method and device |
| CN106164837A (en) * | 2014-04-04 | 2016-11-23 | 电子湾有限公司 | Image evaluation |
| 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 |
| 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 |
| CN108319952A (en) * | 2017-01-16 | 2018-07-24 | 浙江宇视科技有限公司 | A kind of vehicle characteristics extracting method and device |
| 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 |
| CN109102264A (en) * | 2018-08-29 | 2018-12-28 | Oppo广东移动通信有限公司 | A kind of electronics red packet detection method, device and terminal device |
| 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 (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20100054596A1 (en) * | 2008-08-27 | 2010-03-04 | Hankuk University Of Foreign Studies Research And Industry-University Cooperation Foundation | Image segmentation |
| CN101763440A (en) * | 2010-03-26 | 2010-06-30 | 上海交通大学 | Method for filtering searched images |
| US20110229045A1 (en) * | 2010-03-16 | 2011-09-22 | Nec Laboratories America, Inc. | Method and system for image classification |
| 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 |
-
2013
- 2013-01-30 CN CN201310037741.2A patent/CN103106265B/en not_active Expired - Fee Related
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20100054596A1 (en) * | 2008-08-27 | 2010-03-04 | Hankuk University Of Foreign Studies Research And Industry-University Cooperation Foundation | Image segmentation |
| US20110229045A1 (en) * | 2010-03-16 | 2011-09-22 | Nec Laboratories America, Inc. | Method and system for image classification |
| 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 |
|---|
| 张尤赛等: "基于梯度基元聚合矢量的图像检索算法", 《计算机应用研究》 * |
| 王志瑞等: "图像特征提取方法的综述", 《吉首大学学报（自然科学版）》 * |
| 程刚等: "基于结构和纹理特征融合的场景图像分类", 《计算机工程》 * |
Cited By (20)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103530645A (en) * | 2013-09-25 | 2014-01-22 | 北京工商大学 | Texture image classifying method and system based on local binary pattern and Zernike moment |
| 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 |
| US12131342B2 (en) | 2014-04-04 | 2024-10-29 | Ebay Inc | Image evaluation |
| US10176406B2 (en) | 2014-04-04 | 2019-01-08 | Ebay Inc. | Image evaluation |
| CN113792176A (en) * | 2014-04-04 | 2021-12-14 | 电子湾有限公司 | Image evaluation |
| CN106164837A (en) * | 2014-04-04 | 2016-11-23 | 电子湾有限公司 | Image evaluation |
| US11449719B2 (en) | 2014-04-04 | 2022-09-20 | 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 |
| CN105809182A (en) * | 2014-12-31 | 2016-07-27 | 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 |
| CN108319952A (en) * | 2017-01-16 | 2018-07-24 | 浙江宇视科技有限公司 | A kind of vehicle characteristics extracting 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 |
| CN109102264A (en) * | 2018-08-29 | 2018-12-28 | Oppo广东移动通信有限公司 | A kind of electronics red packet detection method, device and terminal device |
| 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 |
Also Published As
| Publication number | Publication date |
|---|---|
| CN103106265B (en) | 2016-10-12 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN103106265A (en) | Method and system of classifying similar images | |
| CN108830188B (en) | Vehicle detection method based on deep learning | |
| CN103514456B (en) | Image classification method and device based on compressed sensing multi-core learning | |
| France et al. | A new approach to automated pollen analysis | |
| CN102054178B (en) | A kind of image of Chinese Painting recognition methods based on local semantic concept | |
| CN107766933B (en) | Visualization method for explaining convolutional neural network | |
| CN103761531B (en) | The sparse coding license plate character recognition method of Shape-based interpolation contour feature | |
| Alidoost et al. | A CNN-based approach for automatic building detection and recognition of roof types using a single aerial image | |
| CN103996018B (en) | Face identification method based on 4DLBP | |
| CN107368787A (en) | A kind of Traffic Sign Recognition algorithm that application is driven towards depth intelligence | |
| CN105956560A (en) | Vehicle model identification method based on pooling multi-scale depth convolution characteristics | |
| CN103984948B (en) | A kind of soft double-deck age estimation method based on facial image fusion feature | |
| CN106022300A (en) | Traffic sign identifying method and traffic sign identifying system based on cascading deep learning | |
| CN103679191B (en) | An automatic fake-licensed vehicle detection method based on static state pictures | |
| CN106408030A (en) | SAR image classification method based on middle lamella semantic attribute and convolution neural network | |
| Yilmaz et al. | A vehicle detection approach using deep learning methodologies | |
| CN105719285A (en) | Pedestrian detection method based on directional chamfering distance characteristics | |
| CN105825233B (en) | A kind of pedestrian detection method based on on-line study random fern classifier | |
| CN101930549A (en) | Second generation curvelet transform-based static human detection method | |
| CN106960176A (en) | A kind of pedestrian's gender identification method based on transfinite learning machine and color characteristic fusion | |
| CN110807485B (en) | Method for fusing two-classification semantic segmentation maps into multi-classification semantic map based on high-resolution remote sensing image | |
| CN103186776A (en) | Human detection method based on multiple features and depth information | |
| CN108985145A (en) | The Opposite direction connection deep neural network model method of small size road traffic sign detection identification | |
| CN114332473A (en) | Object detection method, object detection device, computer equipment, storage medium and program product | |
| CN115984537A (en) | Image processing method and device and related equipment |
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 |