CN108228761A - The customized image search method in support area and device, equipment, medium - Google Patents
The customized image search method in support area and device, equipment, medium Download PDFInfo
- Publication number
- CN108228761A CN108228761A CN201711415215.XA CN201711415215A CN108228761A CN 108228761 A CN108228761 A CN 108228761A CN 201711415215 A CN201711415215 A CN 201711415215A CN 108228761 A CN108228761 A CN 108228761A
- Authority
- CN
- China
- Prior art keywords
- image
- retrieved
- feature
- stored image
- stored
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
- G06F16/58—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
- G06F16/583—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
Abstract
The embodiment of the invention discloses a kind of customized image search method in support area and device, equipment, medium, including:It is searched for from image library based on image to be retrieved and obtains at least one and images match to be retrieved the first image set;At least one self defined area is set to the object to be retrieved in the image to be retrieved;Based on the self defined area of the object to be retrieved, each pre-stored image concentrated to described first image is screened, and obtains the second image set with the images match to be retrieved.The above embodiment of the present invention protrudes the regional area in pending image by self defined area, and the entirety and part for combining image carry out image retrieval, while prominent local feature, ensure that global feature is not ignored.
Description
Technical field
The present invention relates to computer vision technique, especially a kind of customized image search method in support area.
Background technology
With the development of computer vision technique and increasingly ripe, image retrieval has been applied to every field, such as:Base
It is played an important role in the vehicle retrieval system of image in fields such as intelligent transportation, security protections.This kind of system is captured by calculating
Similitude between vehicle pictures and the vehicle pictures of lane database according to similitude from high to low, returns to retrieval knot in an orderly manner
Fruit.
Link is calculated in image similarity, is usually directed to the characterization technique of image.At present, characterization technique is broadly divided into
Two classes:(1) traditional characteristic engineering technology;(2) depth learning technology.
Invention content
The customized image retrieval technologies in a kind of support area provided in an embodiment of the present invention.
One side according to embodiments of the present invention, the customized image search method in a kind of support area provided, packet
It includes:
It is searched for from image library based on image to be retrieved and obtains at least one and images match to be retrieved the first figure
Image set;The image to be retrieved includes at least one object to be retrieved, and described image library includes at least one pre-stored image,
Each pre-stored image includes the object that prestores, and described first image collection includes at least one and image to be retrieved
Matched pre-stored image;
At least one self defined area is set to the object to be retrieved in the image to be retrieved;
Based on the self defined area of the object to be retrieved, each pre-stored image that described first image is concentrated is carried out
Screening, obtains the second image set with the images match to be retrieved, and second image set is treated including at least one with described
Retrieve the matched pre-stored image of the self defined area of object.
In another embodiment based on the above method of the present invention, the figure that prestores described in corresponding to is further included in described image library
Prestore the pre-stored image feature of object as described in;
It is described searched for from image library based on image to be retrieved obtain it is at least one with the images match to be retrieved the
One image set, including:
Using neural network, feature extraction operation is performed to image to be retrieved, obtains corresponding to treating for the object to be retrieved
Retrieval character;
The feature to be retrieved is matched with the pre-stored image feature, based on matching result obtain with it is described to be checked
First image set of rope images match.
It is described to be searched for from image library based on image to be retrieved in another embodiment based on the above method of the present invention
At least one and images match to be retrieved the first image set is obtained, including:
Using neural network, feature extraction operation is performed to image to be retrieved, obtains corresponding to treating for the object to be retrieved
Retrieval character;
Using the neural network, feature extraction operation, acquisition pair are performed to the pre-stored image in described image library respectively
Answer the pre-stored image feature of the object that prestore;
The feature to be retrieved is matched with the pre-stored image feature, based on matching result obtain with it is described to be checked
First image set of rope images match.
It is in another embodiment based on the above method of the present invention, the feature to be retrieved and the pre-stored image is special
Sign is matched, and the first image set with the images match to be retrieved is obtained based on matching result, including:
Calculate the similarity of the feature to be retrieved and the pre-stored image feature;
It is greater than or equal to the first setting value, the pre-stored image feature and the feature to be retrieved in response to the similarity
Match, the first image with the images match to be retrieved is formed based on the corresponding pre-stored image of the pre-stored image feature
Collection;
It is less than first setting value in response to the similarity, the pre-stored image feature and the feature to be retrieved are not
Matching.
In another embodiment based on the above method of the present invention, further include:
The down-sampling rate of the neural network is obtained based on the image to be retrieved and the feature to be retrieved.
In another embodiment based on the above method of the present invention, to described to be retrieved right in the image to be retrieved
As setting at least one self defined area, including:
User setting instruction is received, the object to be retrieved in the image to be retrieved is set based on the setting instruction
Put at least one self defined area.
In another embodiment based on the above method of the present invention, to described to be retrieved in the image to be retrieved
After object sets at least one self defined area, further include:
Based on the coordinate of self defined area in object to be retrieved described in the image zooming-out to be retrieved, obtain described in corresponding to certainly
The coordinate set of definition region.
In another embodiment based on the above method of the present invention, based on to be retrieved described in the image zooming-out to be retrieved
The coordinate of self defined area in object obtains the coordinate set of the corresponding self defined area, including:
Extract the coordinate on self defined area at least one vertex in the image to be retrieved in the object to be retrieved
The width information and elevation information of information and/or the self defined area;
Coordinate information and/or the width information and elevation information based on the vertex form the self defined area
Coordinate set;Each self defined area corresponds to a coordinate set.
In another embodiment based on the above method of the present invention, the feature to be searched includes characteristic pattern to be searched,
Described first image concentrates the corresponding pre-stored image feature of pre-stored image to include characteristics of image figure;
The method of the present invention further includes:
Position of the self defined area in the characteristic pattern is determined based on the coordinate set, is obtained based on the position
Obtain the user-defined feature collection that the self defined area is corresponded in the characteristic pattern to be searched;
And/or self defined area described in the location determination based on the self defined area in the characteristic pattern is described
Mapping position in characteristics of image figure is obtained in the pre-stored image feature based on the mapping position and corresponds to the self-defined area
The mappings characteristics collection in domain.
In another embodiment based on the above method of the present invention, the self-defined area is determined based on the coordinate set
Position of the domain in the characteristic pattern, including:
Using the down-sampling rate, position of the self defined area in the characteristic pattern is obtained based on the coordinate set
It puts.
In another embodiment based on the above method of the present invention, the self-defined area based on the object to be retrieved
Domain, each pre-stored image concentrated to described first image are screened, and obtain the second image with the images match to be retrieved
Collection, including:
Based on the user-defined feature collection and the mappings characteristics collection,
Or based on the self-defining image image set and the mappings characteristics collection and the pending feature and described first image
The corresponding pre-stored image feature of pre-stored image of concentration;
Determine that described first image concentrates the matched pre-stored image of the self defined area with the object to be retrieved;
Second image set is formed based on the matched pre-stored image.
It is special based on the user-defined feature collection and the mapping in another embodiment based on the above method of the present invention
Collection determines that described first image concentrates the matched pre-stored image of the self defined area with the object to be retrieved, including:
Calculate the Regional Similarity between the user-defined feature collection and the mappings characteristics collection;
Using linear weighted function averaging method, obtained based on the Regional Similarity all self-defined in the pending image
Region and the overall similarity of all mapping areas in the pre-stored image;
Determine that described first image concentrates the self-defined area with the object to be retrieved based on the overall similarity
The matched pre-stored image in domain.
In another embodiment based on the above method of the present invention, first figure is determined based on the overall similarity
In image set with the matched pre-stored image of the self defined area of the object to be retrieved, including:
Judge whether the overall similarity is greater than or equal to the second setting value;
It is greater than or equal to the second setting value, the corresponding pre-stored image of the overall similarity in response to the overall similarity
It is matched with the self defined area of the object to be retrieved;
Be less than the second setting value in response to the overall similarity, the corresponding pre-stored image of the overall similarity with it is described
The self defined area of object to be retrieved mismatches.
It is special based on the self-defining image image set and the mapping in another embodiment based on the above method of the present invention
The corresponding characteristics of image of image that collection and the pending feature and described first image are concentrated, determines described first image
The matched pre-stored image of the self defined area with the object to be retrieved is concentrated, including:
Calculate the Regional Similarity between the user-defined feature collection and the mappings characteristics collection;And the feature to be retrieved
With the similarity of the pre-stored image feature;
Using linear weighted function averaging method, the pending image is obtained based on the Regional Similarity and the similarity
With the comprehensive similarity of the pre-stored image;
Determine that described first image concentrates the self-defined area with the object to be retrieved based on the comprehensive similarity
The matched pre-stored image in domain.
In another embodiment based on the above method of the present invention, first figure is determined based on the comprehensive similarity
In image set with the pre-stored image of the images match to be retrieved, including:
Judge whether the comprehensive similarity is greater than or equal to third setting value;
It is greater than or equal to third setting value, the corresponding pre-stored image of the comprehensive similarity in response to the comprehensive similarity
With the images match to be retrieved;
Be less than third setting value in response to the comprehensive similarity, the corresponding pre-stored image of the comprehensive similarity with it is described
Image to be retrieved mismatches.
It is described to be searched for from image library based on image to be retrieved in another embodiment based on the above method of the present invention
After obtaining at least one the first image set with the images match to be retrieved, further include:
Attributive character based on the corresponding pending feature acquisition correspondence of image to be retrieved object to be retrieved;The category
Property feature be used to represent attributive classification of the object to be searched in corresponding attribute.
It is described to be searched for from image library based on image to be retrieved in another embodiment based on the above method of the present invention
At least one and images match to be retrieved the first image set is obtained, including:
It is matched and obtained from described image library based on the corresponding feature to be retrieved of image to be retrieved and the attributive character
At least one image is obtained, the first image set is obtained based on the matched image.
In another embodiment based on the above method of the present invention, based on the corresponding spy to be retrieved of the image to be retrieved
Sign obtains the attributive character of corresponding object to be retrieved, including:
Using Attribute Recognition network, each attribute is based on to the pending feature and carries out tagsort, is distinguished
The attributive character of corresponding each attribute;The Attribute Recognition network is trained by sample characteristics, and the sample characteristics are labeled with
Mark attributive classification.
According to the other side of the embodiment of the present disclosure, the customized image retrieving apparatus in a kind of support area provided,
Including:
Matching unit, search for for being based on image to be retrieved from image library obtain it is at least one with the image to be retrieved
Matched first image set;The image to be retrieved includes at least one object to be retrieved, and described image library is included at least
One pre-stored image, each pre-stored image include the object that prestores, described first image collection include it is at least one and
The pre-stored image of the images match to be retrieved;
Area setting unit, at least one self-defined to the object setting to be retrieved in the image to be retrieved
Region;
Screening unit for the self defined area based on the object to be retrieved, is concentrated described first image
Each pre-stored image is screened, and obtains the second image set with the images match to be retrieved, and second image set is included extremely
Lack a matched pre-stored image of the self defined area with the object to be retrieved.
In another embodiment based on above device of the present invention, the figure that prestores described in corresponding to is further included in described image library
Prestore the pre-stored image feature of object as described in;
The matching unit, including:
Fisrt feature extraction module for utilizing neural network, performs feature extraction operation to image to be retrieved, obtains pair
Answer the feature to be retrieved of the object to be retrieved;
Characteristic matching module, for the feature to be retrieved to be matched with the pre-stored image feature, based on matching
As a result the first image set with the images match to be retrieved is obtained.
In another embodiment based on above device of the present invention, the matching unit, including:
Fisrt feature extraction module for utilizing neural network, performs feature extraction operation to image to be retrieved, obtains pair
Answer the feature to be retrieved of the object to be retrieved;
Second feature extraction module for utilizing the neural network, is respectively held the pre-stored image in described image library
Row feature extraction operation, corresponded to described in prestore the pre-stored image feature of object;
Characteristic matching module, for the feature to be retrieved to be matched with the pre-stored image feature, based on matching
As a result the first image set with the images match to be retrieved is obtained.
In another embodiment based on above device of the present invention, the characteristic matching module, specifically for calculating institute
State the similarity of feature to be retrieved and the pre-stored image feature;
It is greater than or equal to the first setting value, the pre-stored image feature and the feature to be retrieved in response to the similarity
Match, the first image with the images match to be retrieved is formed based on the corresponding pre-stored image of the pre-stored image feature
Collection;It is less than first setting value in response to the similarity, the pre-stored image feature is mismatched with the feature to be retrieved.
In another embodiment based on above device of the present invention, further include:
Down-sampling rate unit, for obtaining the neural network based on the image to be retrieved and the feature to be retrieved
Down-sampling rate.
In another embodiment based on above device of the present invention, the area setting unit is used specifically for receiving
Family setting instruction sets at least one make by oneself based on the setting instruction to the object to be retrieved in the image to be retrieved
Adopted region.
In another embodiment based on above device of the present invention, further include:
Coordinate extraction unit, for the seat based on self defined area in object to be retrieved described in the image zooming-out to be retrieved
Mark obtains the coordinate set of the corresponding self defined area.
In another embodiment based on above device of the present invention, the coordinate extraction unit, including:
Information extraction unit, for extracting the self defined area in the object to be retrieved in the image to be retrieved extremely
The coordinate information on a few vertex and/or the width information and elevation information of the self defined area;
Gather determination unit, for the coordinate information based on the vertex and/or the width information and elevation information structure
Into the coordinate set of the self defined area;Each self defined area corresponds to a coordinate set.
In another embodiment based on above device of the present invention, the feature to be searched includes characteristic pattern to be searched,
Described first image concentrates the corresponding pre-stored image feature of pre-stored image to include characteristics of image figure;
Apparatus of the present invention further include:
Provincial characteristics unit determines position of the self defined area in the characteristic pattern based on the coordinate set,
The user-defined feature collection that the self defined area is corresponded in the characteristic pattern to be searched is obtained based on the position;
And/or self defined area described in the location determination based on the self defined area in the characteristic pattern is described
Mapping position in characteristics of image figure is obtained in the pre-stored image feature based on the mapping position and corresponds to the self-defined area
The mappings characteristics collection in domain.
In another embodiment based on above device of the present invention, the provincial characteristics unit, specifically for utilizing
Down-sampling rate is stated, position of the self defined area in the characteristic pattern is obtained based on the coordinate set.
In another embodiment based on above device of the present invention, the screening unit, specifically for be based on it is described from
Defined feature collection and the mappings characteristics collection based on the self-defining image image set and the mappings characteristics collection and described are waited to locate
Manage the corresponding pre-stored image feature of pre-stored image that feature and described first image are concentrated;Determine described first image concentration and institute
State the matched pre-stored image of the self defined area of object to be retrieved;Described second is formed based on the matched pre-stored image
Image set.
In another embodiment based on above device of the present invention, the screening unit, including:
First computing module, it is similar for calculating the region between the user-defined feature collection and the mappings characteristics collection
Degree;
Overall similarity module, for utilize linear weighted function averaging method, based on the Regional Similarity obtain described in treat
Handle all self defined areas and the overall similarity of all mapping areas in the pre-stored image in image;
First screening module, for based on the overall similarity determine described first image concentrate with it is described to be retrieved right
The matched pre-stored image of the self defined area of elephant.
In another embodiment based on above device of the present invention, first screening module, specifically for judging
State whether overall similarity is greater than or equal to the second setting value;
It is greater than or equal to the second setting value, the corresponding pre-stored image of the overall similarity in response to the overall similarity
It is matched with the self defined area of the object to be retrieved;
Be less than the second setting value in response to the overall similarity, the corresponding pre-stored image of the overall similarity with it is described
The self defined area of object to be retrieved mismatches.
In another embodiment based on above device of the present invention, the screening unit, including:
Second computing module, it is similar for calculating the region between the user-defined feature collection and the mappings characteristics collection
Degree;And the similarity of the feature to be retrieved and the pre-stored image feature;
Comprehensive similarity module, for utilizing linear weighted function averaging method, based on the Regional Similarity and described similar
Degree obtains the pending image and the comprehensive similarity of the pre-stored image;
Second screening module, for based on the comprehensive similarity determine described first image concentrate with it is described to be retrieved right
The matched pre-stored image of the self defined area of elephant.
In another embodiment based on above device of the present invention, second screening module, specifically for judging
State whether comprehensive similarity is greater than or equal to third setting value;
It is greater than or equal to third setting value, the corresponding pre-stored image of the comprehensive similarity in response to the comprehensive similarity
With the images match to be retrieved;
Be less than third setting value in response to the comprehensive similarity, the corresponding pre-stored image of the comprehensive similarity with it is described
Image to be retrieved mismatches.
In another embodiment based on above device of the present invention, further include:
Template(-let), for the category based on the corresponding pending feature acquisition correspondence of image to be retrieved object to be retrieved
Property feature;The attributive character is used to represent attributive classification of the object to be searched in corresponding attribute.
In another embodiment based on above device of the present invention, the matching unit, specifically for being based on described treat
The retrieval corresponding feature to be retrieved of image and the attributive character match from described image library and obtain at least one image, are based on
The matched image obtains the first image set.
In another embodiment based on above device of the present invention, the template(-let), specifically for being known using attribute
Other network is based on the pending feature each attribute and carries out tagsort, corresponded to the category of each attribute respectively
Property feature;The Attribute Recognition network is trained by sample characteristics, and the sample characteristics are labeled with mark attributive classification.
According to the other side of the embodiment of the present disclosure, a kind of electronic equipment provided, including processor, the processor
Including the customized image retrieving apparatus in support area as described above.
According to the other side of the embodiment of the present disclosure, a kind of electronic equipment provided, including:Memory, for storing
Executable instruction;
And processor, it completes to prop up as described above to perform the executable instruction for communicating with the memory
Hold the image search method of zone custom.
According to the other side of the embodiment of the present disclosure, a kind of computer storage media provided, for storing computer
The instruction that can be read, described instruction, which is performed, performs the customized image search method in support area as described above.
According to the other side of the embodiment of the present disclosure, a kind of computer program provided, including computer-readable code,
When the computer-readable code in equipment when running, the processor execution in the equipment is used to implement support as described above
The instruction of the image search method of zone custom.
Based on the customized image search method in a kind of support area that the above embodiment of the present invention provides, based on to be retrieved
Image is searched for from image library obtains at least one and images match to be retrieved the first image set;It is determined by preliminary search
The substantially similar image with image to be retrieved;At least one self defined area is set to the object to be retrieved in image to be retrieved;
Based on the self defined area of object to be retrieved, each pre-stored image in the first image set is screened, is obtained and figure to be retrieved
As matched second image set, the regional area in pending image is protruded by self defined area, combines image
Entirety and part carry out image retrieval, while prominent local feature, ensure that global feature is not ignored.
Below by drawings and examples, technical scheme of the present invention is described in further detail.
Description of the drawings
The attached drawing of a part for constitution instruction describes the embodiment of the present invention, and is used to explain together with description
The principle of the present invention.
With reference to attached drawing, according to following detailed description, the present invention can be more clearly understood, wherein:
Fig. 1 is the flow chart of the customized image search method one embodiment in support area of the present invention.
Fig. 2 is the structure diagram of the customized image retrieving apparatus one embodiment in support area of the present invention.
Fig. 3 is the structure diagram for realizing the terminal device of the embodiment of the present application or the electronic equipment of server.
Specific embodiment
Carry out the various exemplary embodiments of detailed description of the present invention now with reference to attached drawing.It should be noted that:Unless in addition have
Body illustrates that the unlimited system of component and the positioned opposite of step, numerical expression and the numerical value otherwise illustrated in these embodiments is originally
The range of invention.
Simultaneously, it should be appreciated that for ease of description, the size of the various pieces shown in attached drawing is not according to reality
Proportionate relationship draw.
It is illustrative to the description only actually of at least one exemplary embodiment below, is never used as to the present invention
And its application or any restrictions that use.
Technology, method and apparatus known to person of ordinary skill in the relevant may be not discussed in detail, but suitable
In the case of, the technology, method and apparatus should be considered as part of specification.
It should be noted that:Similar label and letter represents similar terms in following attached drawing, therefore, once a certain Xiang Yi
It is defined in a attached drawing, then in subsequent attached drawing does not need to that it is further discussed.
The embodiment of the present invention can be applied to computer system/server, can be with numerous other general or specialized calculating
System environments or configuration operate together.Suitable for be used together with computer system/server well-known computing system, ring
The example of border and/or configuration includes but not limited to:Personal computer system, server computer system, thin client, thick client
Machine, hand-held or laptop devices, the system based on microprocessor, set-top box, programmable consumer electronics, NetPC Network PC,
Minicomputer system, large computer system and distributed cloud computing technology environment including any of the above described system, etc..
Computer system/server can be in computer system executable instruction (such as journey performed by computer system
Sequence module) general linguistic context under describe.In general, program module can include routine, program, target program, component, logic, number
According to structure etc., they perform specific task or realize specific abstract data type.Computer system/server can be with
Implement in distributed cloud computing environment, in distributed cloud computing environment, task is long-range by what is be linked through a communication network
Manage what equipment performed.In distributed cloud computing environment, program module can be located at the Local or Remote meter for including storage device
It calculates in system storage medium.
In the implementation of the present invention, inventor has found, the prior art has at least the following problems:Based on traditional characteristic
The searching system of engineering technology, due to being too dependent on exploitation when set fixed character, the study of retrieval precision relative depth
Technology is low, and universality is not strong enough in practical applications, and the engineer that the update of feature need be numerous and diverse.Based on deep learning
The searching system of technology, although can be by constantly from data learning, improving performance, in practical applications, similar vehicle
Similitude between example is very close, it is difficult to improve the identification of system by simple on-line study.
Fig. 1 is the flow chart of the customized image search method one embodiment in support area of the present invention.As shown in Figure 1,
The embodiment method includes:
Step 101, searched for from image library based on image to be retrieved obtain it is at least one with images match to be retrieved the
One image set.
Wherein, image to be retrieved includes at least one object to be retrieved, and image library includes at least one pre-stored image,
Each pre-stored image includes the object that prestores, and the first image set includes at least one figure that prestores with images match to be retrieved
Picture, the matching for matching essence and being object to be retrieved between the object that prestores of image to be retrieved and pre-stored image, such as:When to be checked
When rope object is face, matching between image to be retrieved and pre-stored image is exactly the matching between facial image;When to be retrieved
When object is vehicle, matching between image to be retrieved and pre-stored image is exactly the matching between vehicle image.
Specifically, image to be retrieved and pre-stored image are subjected to the phase that matching is normally based between the corresponding feature of image
It is matched like degree, is also based on other modes certainly and is matched, the present embodiment is not limited specific matching way.
Optionally, before being matched, object to be retrieved can also be cut out coming from the image to be retrieved, base
In the object to be retrieved for cutting out acquisition similarity mode is carried out with pre-stored image;Specific cut out can be primarily based on neural network knowledge
Do not go out the position of object to be retrieved in object to be retrieved, cut the object to be retrieved from image to be retrieved based on obtained coordinate
It determines and, such as vehicle image from image to be retrieved is identified and is cut out and is come.
Step 102, at least one self defined area is set to the object to be retrieved in image to be retrieved.
Specifically, in one or more alternative embodiments, self defined area can be according to the setting of user instruct into
Row setting, process include:User setting instruction is received, the object to be retrieved in image to be retrieved is set based on setting instruction
At least one self defined area.
Step 103, the self defined area based on object to be retrieved, screen the first image set in each pre-stored image into
Row screening, obtains the second image set with images match to be retrieved.
Wherein, the second image set includes at least one and object to be retrieved matched pre-stored image of self defined area.
Based on the customized image search method in a kind of support area that the above embodiment of the present invention provides, based on to be retrieved
Image is searched for from image library obtains at least one and images match to be retrieved the first image set;It is determined by preliminary search
The substantially similar image with image to be retrieved;At least one self defined area is set to the object to be retrieved in image to be retrieved;
Based on the self defined area of object to be retrieved, each pre-stored image in the first image set is screened, is obtained and figure to be retrieved
As matched second image set, the regional area in pending image is protruded by self defined area, combines image
Entirety and part carry out image retrieval, while prominent local feature, ensure that global feature is not ignored.
In a specific example of the above-mentioned customized image search method above-described embodiment in support area of the present invention, figure
Pre-stored image feature as further including the object that prestores in corresponding pre-stored image in library;
Operation 101 includes:
Using neural network, feature extraction operation is performed to image to be retrieved, obtains corresponding to the to be retrieved of image to be retrieved
Feature;
Feature to be retrieved with pre-stored image feature is matched, is obtained and images match to be retrieved based on matching result
First image set.
In the present embodiment, when the pre-stored image in image library is in preservation storage, feature extraction is carried out, in matched mistake
Cheng Zhong, directly by characteristics of image and pre-stored image feature into the distance between row vector or angle calculation, to realize feature based
The distance between vector or angle obtain the purpose of the similarity between two images, and specific distance may include:Euclidean distance,
COS distance, Hamming distances etc., angle are usually included angle cosine.
In a specific example of the above-mentioned customized image search method above-described embodiment in support area of the present invention, behaviour
Make 101 to include:
Using neural network, feature extraction operation is performed to image to be retrieved, obtains corresponding to the to be retrieved of object to be retrieved
Feature;
Using neural network, feature extraction operation is performed to the pre-stored image in image library respectively, obtains corresponding prestore pair
The pre-stored image feature of elephant;
Feature to be retrieved with pre-stored image feature is matched, is obtained and images match to be retrieved based on matching result
First image set.
In the present embodiment, when the pre-stored image in image library is in preservation storage, face characteristic extraction is not carried out, is being matched
During, it is necessary first to in image library pre-stored image carry out feature extraction, with realize between feature based vector away from
From or angle obtain the purpose of the similarity between two images, specific distance may include:Euclidean distance, Hamming distances etc.,
Angle is usually included angle cosine.
In a specific example of the above-mentioned customized image search method the various embodiments described above in support area of the present invention,
Feature to be retrieved with pre-stored image feature is matched, the first image with images match to be retrieved is obtained based on matching result
Collection, including:
Calculate the similarity of feature to be retrieved and pre-stored image feature;
It is greater than or equal to the first setting value in response to similarity, pre-stored image feature matches with feature to be retrieved, is based on
The corresponding pre-stored image of pre-stored image feature forms the first image set with images match to be retrieved;
It is less than the first setting value in response to similarity, pre-stored image feature is mismatched with feature to be retrieved.
In the present embodiment, judge whether two features match based on the similarity between characteristics of image, when similarity is big
When the first setting value, it is believed that the pre-stored image and image preliminary matches to be retrieved, by pre-stored image deposit the
One image set.
As an optional example, when image to be retrieved is vehicle image, by the vehicle pictures in image library, instruction is utilized
Its feature vector of the network calculations perfected;Signified network is the convolutional neural networks that can predict a variety of vehicle attributes, if multiple
Attribute is predicted that is, signified network is these independent multiple independent convolutional Neurals respectively by multiple independent convolutional neural networks respectively
Network;Signified network is replaceable.Signified feature vector, the characteristic pattern for signified last layer of convolutional layer output of network.If
Signified network is the multiple-branching construction network of shared low-level image feature, then signified feature vector is last before connecting branched structure
The characteristic pattern of one layer of convolutional layer output.The dimension of characteristic pattern is (W × H × C), wherein, W, H, C are the width of characteristic pattern, height respectively
And port number.
In a specific example of the above-mentioned customized image search method the various embodiments described above in support area of the present invention,
The down-sampling rate of neural network is obtained based on image to be retrieved and feature to be retrieved.
Down-sampling rate includes lateral down-sampling rate and longitudinal down-sampling rate.
Specifically, it calculates shown in the formula such as formula (1) of the down-sampling rate of neural network and formula (2):
Wherein, rxAnd ryRespectively lateral down-sampling rate and longitudinal down-sampling rate.Based on above-mentioned formula (1) and formula (2),
Wherein wfRepresent the width value of image to be retrieved, hfRepresent the height value of image to be retrieved, w 'fRepresent the spy of neural network output
Levy the width value of figure, h 'fRepresent the height value of the characteristic pattern of neural network output.It can be based on image to be retrieved and feature to be retrieved
Obtain the down-sampling rate of the neural network of the present embodiment use.
Another embodiment of the customized image search method in support area of the present invention, on the basis of the various embodiments described above
On, after operation 102, further include:
Based on the coordinate of self defined area in image zooming-out to be retrieved object to be retrieved, the seat of corresponding self defined area is obtained
Mark set.
The present embodiment can be set to protrude the local feature of object to be retrieved according to user setting in object to be retrieved
At least one self defined area, in order to obtain the corresponding feature of self defined area in order to subsequently be matched, by self-defined
The partial coordinates in region determine the coordinate set of the self defined area.
In a specific example of the above-mentioned customized image search method the various embodiments described above in support area of the present invention,
Based on the coordinate of self defined area in image zooming-out to be retrieved object to be retrieved, the coordinate set of corresponding self defined area is obtained,
Including:
Extract the self defined area in object to be retrieved in image to be retrieved the coordinate information at least one vertex and/
Or the width information and elevation information of self defined area;
Coordinate information and/or width information and elevation information based on vertex form the coordinate set of self defined area;Often
A self defined area corresponds to a coordinate set.
Specifically, the region in the generally rectangular frame of self defined area for the region in rectangle frame, only need to determine one
The elevation information and width information of apex coordinate and rectangle frame are all coordinates that can determine the rectangle frame, pass through all coordinate structures
Into coordinate set corresponding feature set can be obtained from feature to be retrieved.
Optionally, the coordinate set of self defined area can be expressed as formula (3):
Wherein, xi,yi,wi,hiThe left upper apex of the bounding box of i-th of self defined area is horizontal on image respectively to be retrieved
Coordinate, left upper apex ordinate, the width of bounding box, the height of bounding box.
In a specific example of the above-mentioned customized image search method the various embodiments described above in support area of the present invention,
Feature to be searched includes characteristic pattern to be searched, and the corresponding pre-stored image feature of pre-stored image includes characteristics of image in the first image set
Figure;
The method of the present invention further includes:
Position of the self defined area in characteristic pattern is determined based on coordinate set, is obtained in characteristic pattern to be searched based on position
The user-defined feature collection of corresponding self defined area;
And/or location determination self defined area the reflecting in characteristics of image figure based on self defined area in characteristic pattern
Position is penetrated, the mappings characteristics collection that self defined area is corresponded in pre-stored image feature is obtained based on mapping position.
Optionally, position of the self defined area in characteristic pattern can correspond to the figure that prestores in characteristics of image figure
At least one of picture mapping position, in the case of being compared just for specific region, a self defined area position pair
Answer a mapping position in a pre-stored image;It can also be and carry out all positions in self defined area and pre-stored image
It matches, a self defined area position corresponds to multiple mapping positions in a pre-stored image at this time;The size of self defined area
It is identical with the size of mapping position.
Specifically, feature can include characteristic pattern or feature vector, characteristic pattern is characterized as in the present embodiment, by self-defined
The corresponding coordinate set in region, it may be determined that position of the self defined area in characteristic pattern, and then it is corresponding to obtain self defined area
Feature, since characteristics of image figure is identical with the size of characteristic pattern to be searched, self defined area in characteristic pattern to be searched can be with
It is mapped in characteristics of image figure, the position in the corresponding characteristics of image figure of self defined area is determined, based on determining position
Judge the similarity of self defined area in image to be searched and pre-stored image.
In a specific example of the above-mentioned customized image search method the various embodiments described above in support area of the present invention,
Position of the self defined area in characteristic pattern is determined based on coordinate set, including:
Using down-sampling rate, position of the self defined area in characteristic pattern is obtained based on coordinate set.
Specifically, using the down-sampling rate of neural network, position of the coordinate set of self defined area in characteristic pattern is calculated
It puts and is represented byWherein x 'i,y′i,w′i,h′iRespectively the coordinate set of self defined area is in feature
The left upper apex abscissa of the bounding box of figure, left upper apex ordinate, the width of bounding box, the height of bounding box, w 'fRepresent god
The width value of characteristic pattern through network output, h 'fRepresent the height value of the characteristic pattern of neural network output.x′i,y′i,w′i,h′i
It is obtained by the following formula (4), formula (5), formula (6) and formula (7):
In a specific example of the above-mentioned customized image search method the various embodiments described above in support area of the present invention,
Based on the self defined area of object to be retrieved, each pre-stored image in the first image set is screened, is obtained and figure to be retrieved
As matched second image set, including:
Based on user-defined feature collection and mappings characteristics collection,
Or based on the pre-stored image pair in self-defining image image set and mappings characteristics collection and pending feature and the first image set
The pre-stored image feature answered;
Determine the matched pre-stored image of self defined area with object to be retrieved in the first image set;
Second image set is formed based on determining pre-stored image.
In the present embodiment, individually pre-stored image can again be retrieved by each self defined area character pair, also
The corresponding feature of self defined area with image global feature can be combined and pre-stored image is retrieved again, retrieval again is a kind of
The scheme of the accuracy of searching system can be further enhanced, mainly there are two class particular techniques at present:(1) it is to be increased using regional area
By force;(2) it is assisted using the attribute tags of target.However, the deficiency of (1) is at two, first, the setting of fixed regional area is not
Enhancing can centainly be played the role of, second is that the regional area selection technique that can learn, which may not necessarily focus on, really high identification
Region, the reason is that its excessive fitting training data situation about being marked, under true application scenarios, sometimes high identification
Region tends to belong to abnormal conditions (such as scratch, different target can be different), is usually all not included in training data.(2)
Deficiency is that all kinds of identifiable attribute is pre-set, and it is that impossible include all situations to preset, some
The region of high identification is unpredictable.
The present embodiment overcomes the shortcomings that retrieving again in the prior art by self defined area, in the base for retaining global feature
Customized local feature is added on plinth, makes retrieval effectiveness more preferably.
In a specific example of the above-mentioned customized image search method the various embodiments described above in support area of the present invention,
Based on the corresponding characteristics of image of image in self-defining image image set and mappings characteristics collection and pending feature and the first image set,
Determine in the first image set with the matched pre-stored image of the self defined area of object to be retrieved, including:
Calculate the Regional Similarity between user-defined feature collection and mappings characteristics collection;And feature to be retrieved and pre-stored image are special
The similarity of sign;
Using linear weighted function averaging method, pending image and pre-stored image are obtained based on Regional Similarity and similarity
Comprehensive similarity;
Based on comprehensive similarity determine in the first image set with the matched pre-stored image of the self defined area of object to be retrieved.
Specifically, it sets shown in original measuring similarity function such as formula (8):
S=Sim (Xg) formula (8)
Wherein, XgFor the feature vector of a pair of of picture, s is their similarity measure values, and Sim represents to calculate two feature phases
Like the prior art of degree, such as:Euclidean distance, COS distance etc..
The following formula (9) and formula (10) calculating can be based on by calculating comprehensive similarity using linear weighted function averaging method:
Wherein, XgGlobal characteristics for a pair of of picture are vectorial,For their corresponding i-th of regional areas feature to
Amount, αiThe impact factor of i-th of local features for image can freely be adjusted by user, and λ is auto-adaptive parameter, be met
Constraints:S ' is their similarity measure values;
In a specific example of the above-mentioned customized image search method the various embodiments described above in support area of the present invention,
Pre-stored image in first image set with images match to be retrieved is determined based on comprehensive similarity, including:
Judge whether comprehensive similarity is greater than or equal to third setting value;
In response to comprehensive similarity be greater than or equal to third setting value, the corresponding pre-stored image of comprehensive similarity with it is to be retrieved
Images match;
It is less than third setting value in response to comprehensive similarity, the corresponding pre-stored image of comprehensive similarity and image to be retrieved are not
Matching.
Specifically, similarity calculation can be based on Euclidean distance or COS distance calculates, and be limited by third setting value comprehensive
Similarity is closed, obtains the target image more like with image to be searched.
In a specific example of the above-mentioned customized image search method the various embodiments described above in support area of the present invention,
Based on user-defined feature collection and mappings characteristics collection, determine matched with the self defined area of object to be retrieved in the first image set
Pre-stored image, including:
Calculate the Regional Similarity between user-defined feature collection and mappings characteristics collection;
Using linear weighted function averaging method, based on all self defined areas in the pending image of Regional Similarity acquisition and in advance
Deposit the overall similarity of all mapping areas in image;
Based on overall similarity determine in the first image set with the matched pre-stored image of the self defined area of object to be retrieved.
Specifically, similar to the above embodiments, it is whole similar with formula (10) calculating that formula (9) can also be used in the present embodiment
Degree, only Sim (X in formula at this timeg) it is 0, that is, the similarity of each self defined area is only calculated, is weighted average.
Optionally, it is determined based on overall similarity matched pre- with the self defined area of object to be retrieved in the first image set
Image is deposited, including:
Judge whether overall similarity is greater than or equal to the second setting value;
In response to overall similarity be greater than or equal to the second setting value, the corresponding pre-stored image of overall similarity with it is to be retrieved
The self defined area matching of object;
It is less than the second setting value, the corresponding pre-stored image of overall similarity and object to be retrieved in response to overall similarity
Self defined area mismatches.
The a still further embodiment of the customized image search method in support area of the present invention, on the basis of the various embodiments described above
On, after operation 101, it can also include:
The attributive character of corresponding object to be retrieved is obtained based on the corresponding pending feature of image to be retrieved.
Wherein, attributive character is used to represent attributive classification of the object to be searched in corresponding attribute.
Specifically, it is realized by attributive classification and accurately identifies a variety of image attributes, improve the specific aim of retrieval, propped up
It holds and is configured flexibly retrieval tasks.Such as:The corresponding attribute of vehicle image can include but is not limited to:Vehicle, vehicle body face
Color, Che Pin;The attributive classification of specific corresponding vehicle includes but not limited to:It is car, cargo, small truck, truck, cross-country
Vehicle, bus, minibus;The attributive classification of specific corresponding body color includes but not limited to:Grey, white, red, black,
Blue, green, yellow, brown, purple, pink colour etc.;The attributive classification of specific corresponding vehicle product can include being more than 3000 kinds.
In a specific example of the above-mentioned customized image search method the various embodiments described above in support area of the present invention,
It is searched for from image library based on image to be retrieved and obtains at least one and images match to be retrieved the first image set, including:
It is matched from image library based on the corresponding pending feature of image to be retrieved and attributive character and obtains at least one figure
Picture obtains the first image set based on matched image.
Specifically, pre-stored image is screened first by attributive character, by attributive character and object matching to be retrieved
The matched pre-stored image of conduct, the unmatched matching that will need not carry out next step.
In a specific example of the above-mentioned customized image search method the various embodiments described above in support area of the present invention,
Corresponding attributive character is obtained based on the corresponding pending feature of image to be retrieved, including:
Using Attribute Recognition network, treat processing feature and be based on each attribute progress tagsort, corresponded to each category respectively
The attributive character of property;Attribute Recognition network is trained by sample characteristics, and sample characteristics are labeled with mark attributive classification.
The present embodiment specifically, based on characteristics of image realizes attribute forecast, is all respectively provided with based on every attribute different
Class categories can identify the corresponding attribute institute of the image to be retrieved based on the Attribute Recognition network after training based on characteristics of image
The class categories of category.
The customized image search method in support area of the present invention can be applied to be retrieved with facial image retrieval, vehicle image
The retrieval in the fields of grade.
The process that concrete application is retrieved with vehicle image may include:
1st, input/importing target vehicle image (vehicle image to be retrieved);
2nd, selection vehicle library;
3rd, system automatically retrieval obtains retrieval result (the first image set) for the first time;
4th, user configuration filter condition (setting self defined area).
Select attribute tags;The manual one or more self defined areas of frame choosing;
5th, secondary system is retrieved, and obtains quadratic search result;
6th, output/export quadratic search result;
7th, it is such as still dissatisfied to quadratic search result, " return " can be clicked to the 3rd step treated state, reconfigured
Filter condition.
One of ordinary skill in the art will appreciate that:Realizing all or part of step of above method embodiment can pass through
The relevant hardware of program instruction is completed, and aforementioned program can be stored in a computer read/write memory medium, the program
When being executed, step including the steps of the foregoing method embodiments is performed;And aforementioned storage medium includes:ROM, RAM, magnetic disc or light
The various media that can store program code such as disk.
Fig. 2 is the structure diagram of the customized image retrieving apparatus one embodiment in support area of the present invention.The implementation
The device of example can be used for realizing the above-mentioned each method embodiment of the present invention.As shown in Fig. 2, the device of the embodiment includes:
Matching unit 21, search for for being based on image to be retrieved from image library obtain it is at least one with image to be retrieved
The first image set matched.
Wherein, image to be retrieved includes at least one object to be retrieved, and image library includes at least one pre-stored image,
Each pre-stored image includes the object that prestores, and the first image set includes at least one figure that prestores with images match to be retrieved
Picture.
Area setting unit 22, for setting at least one self defined area to the object to be retrieved in image to be retrieved.
Specifically, in one or more alternative embodiments, self defined area can be according to the setting of user instruct into
Row setting, process include:User setting instruction is received, the object to be retrieved in image to be retrieved is set based on setting instruction
At least one self defined area.
Screening unit 23, for the self defined area based on object to be retrieved, to each pre-stored image in the first image set
It is screened, obtains the second image set with images match to be retrieved.
Wherein, the second image set includes at least one and object to be retrieved matched pre-stored image of self defined area.
Based on the customized image search method in a kind of support area that the above embodiment of the present invention provides, based on to be retrieved
Image is searched for from image library obtains at least one and images match to be retrieved the first image set;It is determined by preliminary search
The substantially similar image with image to be retrieved;At least one self defined area is set to the object to be retrieved in image to be retrieved;
Based on the self defined area of object to be retrieved, each pre-stored image in the first image set is screened, is obtained and figure to be retrieved
As matched second image set, the regional area in pending image is protruded by self defined area, combines image
Entirety and part carry out image retrieval, while prominent local feature, ensure that global feature is not ignored.
In a specific example of the above-mentioned customized image retrieving apparatus above-described embodiment in support area of the present invention, figure
Pre-stored image feature as further including the object that prestores in corresponding pre-stored image in library;
Matching unit 21, including:
Fisrt feature extraction module for utilizing neural network, performs feature extraction operation to image to be retrieved, obtains pair
Answer the feature to be retrieved of object to be retrieved;
Characteristic matching module for feature to be retrieved to be matched with pre-stored image feature, is obtained based on matching result
With the first image set of images match to be retrieved.
In a specific example of the above-mentioned customized image retrieving apparatus above-described embodiment in support area of the present invention,
With unit 21, including:
Fisrt feature extraction module for utilizing neural network, performs feature extraction operation to image to be retrieved, obtains pair
Answer the feature to be retrieved of object to be retrieved;
Second feature extraction module for utilizing neural network, performs feature to the pre-stored image in image library respectively and carries
Extract operation obtains the pre-stored image feature of the corresponding object that prestores;
Characteristic matching module for feature to be retrieved to be matched with pre-stored image feature, is obtained based on matching result
With the first image set of images match to be retrieved.
In a specific example of the above-mentioned customized image retrieving apparatus the various embodiments described above in support area of the present invention,
Characteristic matching module, specifically for calculating the similarity of feature to be retrieved and pre-stored image feature;
It is greater than or equal to the first setting value in response to similarity, pre-stored image feature matches with feature to be retrieved, is based on
The corresponding pre-stored image of pre-stored image feature forms the first image set with images match to be retrieved;
It is less than the first setting value in response to similarity, pre-stored image feature is mismatched with feature to be retrieved.
In a specific example of the above-mentioned customized image retrieving apparatus above-described embodiment in support area of the present invention, also
Including:
Down-sampling rate unit, for obtaining the down-sampling rate of neural network based on image to be retrieved and feature to be retrieved.
Another embodiment of the customized image retrieving apparatus in support area of the present invention, on the basis of the various embodiments described above
On, it further includes:
Coordinate extraction unit for the coordinate based on self defined area in image zooming-out to be retrieved object to be retrieved, obtains
The coordinate set of corresponding self defined area.
The present embodiment can be set to protrude the local feature of object to be retrieved according to user setting in object to be retrieved
At least one self defined area, in order to obtain the corresponding feature of self defined area in order to subsequently be matched, by self-defined
The partial coordinates in region determine the coordinate set of the self defined area.
In a specific example of the above-mentioned customized image retrieving apparatus the various embodiments described above in support area of the present invention,
Coordinate extraction unit, including:
Information extraction unit, for extracting at least one top in image to be retrieved of the self defined area in object to be retrieved
The coordinate information of point and/or the width information and elevation information of self defined area;
Gather determination unit, formed for the coordinate information based on vertex and/or width information and elevation information self-defined
The coordinate set in region;Each self defined area corresponds to a coordinate set.
In a specific example of the above-mentioned customized image retrieving apparatus the various embodiments described above in support area of the present invention,
Feature to be searched includes characteristic pattern to be searched, and the corresponding pre-stored image feature of pre-stored image includes characteristics of image in the first image set
Figure;
Apparatus of the present invention further include:
Provincial characteristics unit determines position of the self defined area in characteristic pattern based on coordinate set, is obtained based on position
The user-defined feature collection of self defined area is corresponded in characteristic pattern to be searched;
And/or location determination self defined area the reflecting in characteristics of image figure based on self defined area in characteristic pattern
Position is penetrated, the mappings characteristics collection that self defined area is corresponded in pre-stored image feature is obtained based on mapping position.
In a specific example of the above-mentioned customized image retrieving apparatus the various embodiments described above in support area of the present invention,
Specifically for utilizing down-sampling rate, position of the self defined area in characteristic pattern is obtained based on coordinate set for provincial characteristics unit.
In a specific example of the above-mentioned customized image retrieving apparatus the various embodiments described above in support area of the present invention,
Screening unit 23, specifically for being based on user-defined feature collection and mappings characteristics collection or based on self-defining image image set and mappings characteristics
The corresponding pre-stored image feature of pre-stored image in collection and pending feature and the first image set;Determine in the first image set with
The matched pre-stored image of self defined area of object to be retrieved;Second image set is formed based on matched pre-stored image.In this hair
In one specific example of the bright customized image retrieving apparatus the various embodiments described above in above-mentioned support area, screening unit 23, packet
It includes:
First computing module, for calculating the Regional Similarity between user-defined feature collection and mappings characteristics collection;
For utilizing linear weighted function averaging method, pending image is obtained based on Regional Similarity for overall similarity module
In in all self defined areas and pre-stored image all mapping areas overall similarity;
First screening module, for determining the self-defined area in the first image set with object to be retrieved based on overall similarity
The matched pre-stored image in domain.
In a specific example of the above-mentioned customized image retrieving apparatus the various embodiments described above in support area of the present invention,
First screening module, specifically for judging whether overall similarity is greater than or equal to the second setting value;
In response to overall similarity be greater than or equal to the second setting value, the corresponding pre-stored image of overall similarity with it is to be retrieved
The self defined area matching of object;
It is less than the second setting value, the corresponding pre-stored image of overall similarity and object to be retrieved in response to overall similarity
Self defined area mismatches.
In a specific example of the above-mentioned customized image retrieving apparatus the various embodiments described above in support area of the present invention,
Screening unit 23, including:
Second computing module, for calculating the Regional Similarity between user-defined feature collection and mappings characteristics collection;It is and to be checked
Suo Tezheng and the similarity of pre-stored image feature;
Comprehensive similarity module for utilizing linear weighted function averaging method, is treated based on Regional Similarity and similarity
Handle the comprehensive similarity of image and pre-stored image;
Second screening module, for determining the self-defined area in the first image set with object to be retrieved based on comprehensive similarity
The matched pre-stored image in domain.
In a specific example of the above-mentioned customized image retrieving apparatus the various embodiments described above in support area of the present invention,
Second screening module, specifically for judging whether comprehensive similarity is greater than or equal to third setting value;
In response to comprehensive similarity be greater than or equal to third setting value, the corresponding pre-stored image of comprehensive similarity with it is to be retrieved
Images match;
It is less than third setting value in response to comprehensive similarity, the corresponding pre-stored image of comprehensive similarity and image to be retrieved are not
Matching.
The a still further embodiment of the customized image retrieving apparatus in support area of the present invention, on the basis of the various embodiments described above
On, it further includes:
Template(-let), it is special for obtaining the attribute of corresponding object to be retrieved based on the corresponding pending feature of image to be retrieved
Sign.
Wherein, attributive character is used to represent attributive classification of the object to be searched in corresponding attribute.
Specifically, it is realized by attributive classification and accurately identifies a variety of image attributes, improve the specific aim of retrieval, propped up
It holds and is configured flexibly retrieval tasks.Such as:The corresponding attribute of vehicle image can include but is not limited to:Vehicle, vehicle body face
Color, Che Pin;The attributive classification of specific corresponding vehicle includes but not limited to:It is car, cargo, small truck, truck, cross-country
Vehicle, bus, minibus;The attributive classification of specific corresponding body color includes but not limited to:Grey, white, red, black,
Blue, green, yellow, brown, purple, pink colour etc.;The attributive classification of specific corresponding vehicle product can include being more than 3000 kinds.
In a specific example of the above-mentioned customized image retrieving apparatus the various embodiments described above in support area of the present invention,
Matching unit 21 is matched from image library and is obtained specifically for being based on the corresponding feature to be retrieved of image to be retrieved and attributive character
At least one image obtains the first image set based on matched image.
In a specific example of the above-mentioned customized image retrieving apparatus the various embodiments described above in support area of the present invention,
Template(-let) specifically for utilizing Attribute Recognition network, treats processing feature and is based on each attribute progress tagsort, distinguished
The attributive character of corresponding each attribute;Attribute Recognition network is trained by sample characteristics, and sample characteristics are labeled with mark attributive classification.
One side according to embodiments of the present invention, a kind of electronic equipment provided, including processor, processor includes this
Invent the customized image retrieving apparatus in support area of any of the above-described embodiment.
One side according to embodiments of the present invention, a kind of electronic equipment provided, including:Memory, can for storing
Execute instruction;
And processor, for communicating with memory, to perform executable instruction, support area is made by oneself thereby completing the present invention
The operation of any of the above-described embodiment of image search method of justice.
A kind of one side according to embodiments of the present invention, the computer storage media provided, can for storing computer
The instruction of reading, instruction, which is performed, performs any of the above-described embodiment of the customized image search method in support area of the present invention
Operation.
One side according to embodiments of the present invention, a kind of computer program provided, including computer-readable code, when
For computer-readable code when being run in equipment, it is self-defined that the processor execution in the equipment is used to implement support area of the present invention
Image search method any one embodiment instruction.
The embodiment of the present invention additionally provides a kind of electronic equipment, such as can be mobile terminal, personal computer (PC), put down
Plate computer, server etc..Below with reference to Fig. 3, it illustrates suitable for being used for realizing the terminal device of the embodiment of the present application or service
The structure diagram of the electronic equipment 300 of device:As shown in figure 3, computer system 300 includes one or more processors, communication
Portion etc., one or more of processors are for example:One or more central processing unit (CPU) 301 and/or one or more
Image processor (GPU) 313 etc., processor can according to the executable instruction being stored in read-only memory (ROM) 302 or
From the executable instruction that storage section 308 is loaded into random access storage device (RAM) 303 perform various appropriate actions and
Processing.Communication unit 312 may include but be not limited to network interface card, and the network interface card may include but be not limited to IB (Infiniband) network interface card.
Processor can communicate with read-only memory 302 and/or random access storage device 330 to perform executable instruction,
It is connected by bus 304 with communication unit 312 and is communicated through communication unit 312 with other target devices, is implemented so as to complete the application
Example provide the corresponding operation of any one method, for example, searched for from image library based on image to be retrieved obtain it is at least one and
First image set of images match to be retrieved;At least one self defined area is set to the object to be retrieved in image to be retrieved;
The self defined area based on object to be retrieved, each pre-stored image screened in the first image set are screened, obtain and treat
Retrieve the second image set of images match.
In addition, in RAM 303, it can also be stored with various programs and data needed for device operation.CPU301、ROM302
And RAM303 is connected with each other by bus 304.In the case where there is RAM303, ROM302 is optional module.RAM303 is stored
Executable instruction is written in executable instruction into ROM302 at runtime, and it is above-mentioned logical that executable instruction performs processor 301
The corresponding operation of letter method.Input/output (I/O) interface 305 is also connected to bus 304.Communication unit 312 can be integrally disposed,
It may be set to be with multiple submodule (such as multiple IB network interface cards), and in bus link.
I/O interfaces 305 are connected to lower component:Importation 306 including keyboard, mouse etc.;It is penetrated including such as cathode
The output par, c 307 of spool (CRT), liquid crystal display (LCD) etc. and loud speaker etc.;Storage section 308 including hard disk etc.;
And the communications portion 309 of the network interface card including LAN card, modem etc..Communications portion 309 via such as because
The network of spy's net performs communication process.Driver 310 is also according to needing to be connected to I/O interfaces 305.Detachable media 311, such as
Disk, CD, magneto-optic disk, semiconductor memory etc. are mounted on driver 310, as needed in order to be read from thereon
Computer program be mounted into storage section 308 as needed.
Need what is illustrated, framework as shown in Figure 3 is only a kind of optional realization method, can root during concrete practice
The component count amount and type of above-mentioned Fig. 3 are selected, are deleted, increased or replaced according to actual needs;It is set in different function component
Put, can also be used it is separately positioned or integrally disposed and other implementations, such as GPU and CPU separate setting or can be by GPU collection
Into on CPU, communication unit separates setting, can also be integrally disposed on CPU or GPU, etc..These interchangeable embodiments
Each fall within protection domain disclosed by the invention.
Particularly, in accordance with an embodiment of the present disclosure, it may be implemented as computer above with reference to the process of flow chart description
Software program.For example, embodiment of the disclosure includes a kind of computer program product, it is machine readable including being tangibly embodied in
Computer program on medium, computer program are included for the program code of the method shown in execution flow chart, program code
It may include the corresponding instruction of corresponding execution method and step provided by the embodiments of the present application, for example, being based on image to be retrieved from image
Search obtains at least one the first image set with images match to be retrieved in library;Object to be retrieved in image to be retrieved is set
Put at least one self defined area;The self defined area based on object to be retrieved screens respectively prestoring in the first image set
Image is screened, and obtains the second image set with images match to be retrieved.In such embodiments, which can
To be downloaded and installed from network by communications portion 309 and/or be mounted from detachable media 311.In the computer journey
When sequence is performed by central processing unit (CPU) 301, the above-mentioned function of being limited in the present processes is performed.
Methods and apparatus of the present invention, equipment may be achieved in many ways.For example, software, hardware, firmware can be passed through
Or any combinations of software, hardware, firmware realize methods and apparatus of the present invention, equipment.The step of for method
Sequence is stated merely to illustrate, the step of method of the invention is not limited to sequence described in detail above, unless with other
Mode illustrates.In addition, in some embodiments, the present invention can be also embodied as recording program in the recording medium, this
A little programs include being used to implement machine readable instructions according to the method for the present invention.Thus, the present invention also covering stores to hold
The recording medium of the program of row according to the method for the present invention.
Description of the invention provides for the sake of example and description, and is not exhaustively or will be of the invention
It is limited to disclosed form.Many modifications and variations are obvious for the ordinary skill in the art.It selects and retouches
It states embodiment and is to more preferably illustrate the principle of the present invention and practical application, and those of ordinary skill in the art is enable to manage
The solution present invention is so as to design the various embodiments with various modifications suitable for special-purpose.
Claims (10)
1. a kind of customized image search method in support area, which is characterized in that including:
It is searched for from image library based on image to be retrieved and obtains at least one and images match to be retrieved the first image set;
The image to be retrieved includes at least one object to be retrieved, and described image library includes at least one pre-stored image, each
The pre-stored image includes the object that prestores, and described first image collection includes at least one and images match to be retrieved
Pre-stored image;
At least one self defined area is set to the object to be retrieved in the image to be retrieved;
Based on the self defined area of the object to be retrieved, each pre-stored image concentrated to described first image is sieved
Choosing, obtain with the second image set of the images match to be retrieved, second image set include it is at least one with it is described to be checked
The matched pre-stored image of the self defined area of rope object.
2. it according to the method described in claim 1, it is characterized in that, is further included in described image library in the corresponding pre-stored image
The pre-stored image feature of the object that prestores;
Described searched for from image library based on image to be retrieved obtains at least one and images match to be retrieved the first figure
Image set, including:
Using neural network, feature extraction operation is performed to image to be retrieved, obtains corresponding to the to be retrieved of the object to be retrieved
Feature;
The feature to be retrieved with the pre-stored image feature is matched, is obtained and the figure to be retrieved based on matching result
As matched first image set.
3. according to the method described in claim 1, it is characterized in that, described searched for from image library based on image to be retrieved is obtained
At least one the first image set with the images match to be retrieved, including:
Using neural network, feature extraction operation is performed to image to be retrieved, obtains corresponding to the to be retrieved of the object to be retrieved
Feature;
Using the neural network, feature extraction operation is performed to the pre-stored image in described image library respectively, obtains corresponding institute
State the pre-stored image feature for the object that prestores;
The feature to be retrieved with the pre-stored image feature is matched, is obtained and the figure to be retrieved based on matching result
As matched first image set.
4. according to the method in claim 2 or 3, which is characterized in that the feature to be retrieved and the pre-stored image is special
Sign is matched, and the first image set with the images match to be retrieved is obtained based on matching result, including:
Calculate the similarity of the feature to be retrieved and the pre-stored image feature;
It is greater than or equal to the first setting value, the pre-stored image feature and the feature phase to be retrieved in response to the similarity
Match, the first image set with the images match to be retrieved is formed based on the corresponding pre-stored image of the pre-stored image feature;
It is less than first setting value in response to the similarity, the pre-stored image feature and the feature to be retrieved are not
Match.
5. according to any methods of claim 2-4, which is characterized in that further include:
The down-sampling rate of the neural network is obtained based on the image to be retrieved and the feature to be retrieved.
6. a kind of customized image retrieving apparatus in support area, which is characterized in that including:
Matching unit, search for for being based on image to be retrieved from image library obtain it is at least one with the images match to be retrieved
The first image set;The image to be retrieved includes at least one object to be retrieved, and described image library includes at least one
Pre-stored image, each pre-stored image include the object that prestores, described first image collection including it is at least one with it is described
The pre-stored image of images match to be retrieved;
Area setting unit, for setting at least one self-defined area to the object to be retrieved in the image to be retrieved
Domain;
Screening unit for the self defined area based on the object to be retrieved, is concentrated described first image each pre-
It deposits image to be screened, obtains the second image set with the images match to be retrieved, second image set includes at least one
A matched pre-stored image of the self defined area with the object to be retrieved.
7. a kind of electronic equipment, which is characterized in that including processor, the processor includes the support area described in claim 6
The customized image retrieving apparatus in domain.
8. a kind of electronic equipment, which is characterized in that including:Memory, for storing executable instruction;
And processor, for communicating to perform the executable instruction so as to complete claim 1 to 5 times with the memory
One customized image search method in support area of meaning.
9. a kind of computer storage media, for storing computer-readable instruction, which is characterized in that described instruction is performed
When perform claim require 1 to 5 any one described in the customized image search method in support area.
10. a kind of computer program, including computer-readable code, which is characterized in that when the computer-readable code is being set
During standby upper operation, the processor execution in the equipment be used to implement claim 1 to 5 any one described in support area make by oneself
The instruction of the image search method of justice.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201711415215.XA CN108228761B (en) | 2017-12-21 | 2017-12-21 | Image retrieval method and device supporting region customization, equipment and medium |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201711415215.XA CN108228761B (en) | 2017-12-21 | 2017-12-21 | Image retrieval method and device supporting region customization, equipment and medium |
Publications (2)
Publication Number | Publication Date |
---|---|
CN108228761A true CN108228761A (en) | 2018-06-29 |
CN108228761B CN108228761B (en) | 2021-03-23 |
Family
ID=62648675
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201711415215.XA Active CN108228761B (en) | 2017-12-21 | 2017-12-21 | Image retrieval method and device supporting region customization, equipment and medium |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108228761B (en) |
Cited By (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109165620A (en) * | 2018-09-04 | 2019-01-08 | 张晓飞 | A kind of detection method of electric vehicle, system and terminal device |
CN109446364A (en) * | 2018-10-23 | 2019-03-08 | 北京旷视科技有限公司 | Capture search method, image processing method, device, equipment and storage medium |
CN110059212A (en) * | 2019-03-16 | 2019-07-26 | 平安科技(深圳)有限公司 | Image search method, device, equipment and computer readable storage medium |
CN110532964A (en) * | 2019-08-30 | 2019-12-03 | 广东小天才科技有限公司 | Page number recognition methods and device, reading machine people, computer readable storage medium |
CN110851640A (en) * | 2018-07-24 | 2020-02-28 | 杭州海康威视数字技术股份有限公司 | Image searching method, device and system |
CN111078924A (en) * | 2018-10-18 | 2020-04-28 | 深圳云天励飞技术有限公司 | Image retrieval method, device, terminal and storage medium |
CN111126384A (en) * | 2019-12-12 | 2020-05-08 | 创新奇智(青岛)科技有限公司 | Commodity classification system and method based on feature fusion |
CN111783836A (en) * | 2020-06-04 | 2020-10-16 | 北京思特奇信息技术股份有限公司 | Remote store patrol method and device |
WO2023272659A1 (en) * | 2021-06-30 | 2023-01-05 | 东莞市小精灵教育软件有限公司 | Method and apparatus for recognizing cover image, storage medium, and recognition device |
Citations (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101030222A (en) * | 2007-03-22 | 2007-09-05 | 华为技术有限公司 | Model searching device and method |
CN101369316A (en) * | 2008-07-09 | 2009-02-18 | 东华大学 | Image characteristics extraction method based on global and local structure amalgamation |
CN101639858A (en) * | 2009-08-21 | 2010-02-03 | 深圳创维数字技术股份有限公司 | Image search method based on target area matching |
US20100195914A1 (en) * | 2009-02-02 | 2010-08-05 | Michael Isard | Scalable near duplicate image search with geometric constraints |
CN102542058A (en) * | 2011-12-29 | 2012-07-04 | 天津大学 | Hierarchical landmark identification method integrating global visual characteristics and local visual characteristics |
US20130124514A1 (en) * | 2011-03-10 | 2013-05-16 | International Business Machines Corporaiton | Hierarchical ranking of facial attributes |
CN104239898A (en) * | 2014-09-05 | 2014-12-24 | 浙江捷尚视觉科技股份有限公司 | Method for carrying out fast vehicle comparison and vehicle type recognition at tollgate |
US20170140225A1 (en) * | 2015-11-13 | 2017-05-18 | Ricoh Company, Ltd. | Information processing device, information processing method, and information processing system |
US20170193329A1 (en) * | 2016-01-04 | 2017-07-06 | Bank Of America Corporation | Automatic Image Duplication Identification |
CN107145487A (en) * | 2016-03-01 | 2017-09-08 | 深圳中兴力维技术有限公司 | Image search method and device |
-
2017
- 2017-12-21 CN CN201711415215.XA patent/CN108228761B/en active Active
Patent Citations (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101030222A (en) * | 2007-03-22 | 2007-09-05 | 华为技术有限公司 | Model searching device and method |
CN101369316A (en) * | 2008-07-09 | 2009-02-18 | 东华大学 | Image characteristics extraction method based on global and local structure amalgamation |
US20100195914A1 (en) * | 2009-02-02 | 2010-08-05 | Michael Isard | Scalable near duplicate image search with geometric constraints |
CN101639858A (en) * | 2009-08-21 | 2010-02-03 | 深圳创维数字技术股份有限公司 | Image search method based on target area matching |
US20130124514A1 (en) * | 2011-03-10 | 2013-05-16 | International Business Machines Corporaiton | Hierarchical ranking of facial attributes |
CN102542058A (en) * | 2011-12-29 | 2012-07-04 | 天津大学 | Hierarchical landmark identification method integrating global visual characteristics and local visual characteristics |
CN104239898A (en) * | 2014-09-05 | 2014-12-24 | 浙江捷尚视觉科技股份有限公司 | Method for carrying out fast vehicle comparison and vehicle type recognition at tollgate |
US20170140225A1 (en) * | 2015-11-13 | 2017-05-18 | Ricoh Company, Ltd. | Information processing device, information processing method, and information processing system |
US20170193329A1 (en) * | 2016-01-04 | 2017-07-06 | Bank Of America Corporation | Automatic Image Duplication Identification |
CN107145487A (en) * | 2016-03-01 | 2017-09-08 | 深圳中兴力维技术有限公司 | Image search method and device |
Non-Patent Citations (2)
Title |
---|
熊昌镇、单艳梅、郭芬红: ""结合主体检测的图像检索方法"", 《光学精密工程》 * |
赵宏伟、陈霄、石景海、马凌蛟: ""综合颜色和形状特征的交通标志图像检索算法"", 《吉林大学学报(工学版)》 * |
Cited By (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110851640A (en) * | 2018-07-24 | 2020-02-28 | 杭州海康威视数字技术股份有限公司 | Image searching method, device and system |
CN109165620A (en) * | 2018-09-04 | 2019-01-08 | 张晓飞 | A kind of detection method of electric vehicle, system and terminal device |
CN109165620B (en) * | 2018-09-04 | 2022-10-11 | 张晓飞 | Detection method and system of electric vehicle and terminal equipment |
CN111078924A (en) * | 2018-10-18 | 2020-04-28 | 深圳云天励飞技术有限公司 | Image retrieval method, device, terminal and storage medium |
CN109446364A (en) * | 2018-10-23 | 2019-03-08 | 北京旷视科技有限公司 | Capture search method, image processing method, device, equipment and storage medium |
CN110059212A (en) * | 2019-03-16 | 2019-07-26 | 平安科技(深圳)有限公司 | Image search method, device, equipment and computer readable storage medium |
WO2020186777A1 (en) * | 2019-03-16 | 2020-09-24 | 平安科技(深圳)有限公司 | Image retrieval method, apparatus and device, and computer-readable storage medium |
CN110532964A (en) * | 2019-08-30 | 2019-12-03 | 广东小天才科技有限公司 | Page number recognition methods and device, reading machine people, computer readable storage medium |
CN111126384A (en) * | 2019-12-12 | 2020-05-08 | 创新奇智(青岛)科技有限公司 | Commodity classification system and method based on feature fusion |
CN111783836A (en) * | 2020-06-04 | 2020-10-16 | 北京思特奇信息技术股份有限公司 | Remote store patrol method and device |
WO2023272659A1 (en) * | 2021-06-30 | 2023-01-05 | 东莞市小精灵教育软件有限公司 | Method and apparatus for recognizing cover image, storage medium, and recognition device |
Also Published As
Publication number | Publication date |
---|---|
CN108228761B (en) | 2021-03-23 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108228761A (en) | The customized image search method in support area and device, equipment, medium | |
CN104572804B (en) | A kind of method and its system of video object retrieval | |
CN108229479A (en) | The training method and device of semantic segmentation model, electronic equipment, storage medium | |
CN103577475B (en) | A kind of picture mechanized classification method, image processing method and its device | |
CN108920580A (en) | Image matching method, device, storage medium and terminal | |
CN106446150B (en) | A kind of method and device of vehicle precise search | |
CA3154025A1 (en) | Interactive behavior recognizing method, device, computer equipment and storage medium | |
CN108229341A (en) | Sorting technique and device, electronic equipment, computer storage media, program | |
CN108229297A (en) | Face identification method and device, electronic equipment, computer storage media | |
CN109165645A (en) | A kind of image processing method, device and relevant device | |
CN108427927A (en) | Target recognition methods and device, electronic equipment, program and storage medium again | |
CN108230082A (en) | The recommendation method and apparatus of collocation dress ornament, electronic equipment, storage medium | |
CN102034107B (en) | Unhealthy image differentiating method based on robust visual attention feature and sparse representation | |
CN108268527B (en) | A method of detection land use pattern variation | |
CN105005761B (en) | A kind of full-color high resolution remote sensing images Approach for road detection of combination significance analysis | |
CN103617413B (en) | Method for identifying object in image | |
CN106096542A (en) | Image/video scene recognition method based on range prediction information | |
CN110136198A (en) | Image processing method and its device, equipment and storage medium | |
Weber et al. | Spatial and spectral morphological template matching | |
CN108229418A (en) | Human body critical point detection method and apparatus, electronic equipment, storage medium and program | |
CN108491872A (en) | Target recognition methods and device, electronic equipment, program and storage medium again | |
CN110457677A (en) | Entity-relationship recognition method and device, storage medium, computer equipment | |
Liu et al. | Occlusion‐robust traffic sign detection via cascaded colour cubic feature | |
CN109766818A (en) | Pupil center's localization method and system, computer equipment and readable storage medium storing program for executing | |
CN111340782B (en) | Image marking method, device and system |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |