CN110263198A - A kind of search method and device - Google Patents
A kind of search method and device Download PDFInfo
- Publication number
- CN110263198A CN110263198A CN201910567616.XA CN201910567616A CN110263198A CN 110263198 A CN110263198 A CN 110263198A CN 201910567616 A CN201910567616 A CN 201910567616A CN 110263198 A CN110263198 A CN 110263198A
- Authority
- CN
- China
- Prior art keywords
- word
- distance
- matching
- node
- target
- 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.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 98
- 238000007689 inspection Methods 0.000 claims description 4
- 239000000284 extract Substances 0.000 claims description 3
- 235000013399 edible fruits Nutrition 0.000 claims description 2
- 238000010586 diagram Methods 0.000 description 16
- 238000005516 engineering process Methods 0.000 description 8
- 238000013507 mapping Methods 0.000 description 7
- 238000012549 training Methods 0.000 description 5
- 241000406668 Loxodonta cyclotis Species 0.000 description 3
- 230000006870 function Effects 0.000 description 3
- 238000003672 processing method Methods 0.000 description 2
- 230000001174 ascending effect Effects 0.000 description 1
- 239000003086 colorant Substances 0.000 description 1
- 230000005611 electricity Effects 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 239000003973 paint Substances 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
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
- G06F16/51—Indexing; Data structures therefor; Storage structures
-
- 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
- G06F16/53—Querying
- G06F16/532—Query formulation, e.g. graphical querying
Abstract
Search method provided by the invention and device, determine the matching word of object to be processed, and calculate the distance parameter of preset candidate target, and distance parameter is finally met to the candidate target of preset condition, the search result as object to be processed.Wherein, the distance parameter of any one candidate target is determined according to the corresponding target range of matching word, and target range is target word in the preset equivalent of matching word at a distance from the matching word, target word be include the word in candidate target.As it can be seen that search method provided by the present application, it is no longer necessary to the word in the matching word and candidate target in object to be processed be calculated distance one by one, and only need to determine distance parameter according to target range.And equivalent at a distance from matching word (namely target word is at a distance from matching word) precalculates to obtain, the target word in equivalent need to be only obtained in retrieving, and the target word can calculate the distance parameter of candidate target at a distance from matching word, to improve the speed of retrieval.
Description
Technical field
The present invention relates to information discriminating technology field, in particular to a kind of search method and equipment.
Background technique
Attention with parent to early education, early education electronic product is more more and more universal, a kind of common early education class electricity
Sub- product is to draw this interpreter, function be scan that user currently reads draw this page, and use image retrieval technologies are in sample
The sample image that this page matches is retrieved and drawn in image library, when retrieving and drawing the sample image that this page matches, is played
The explanation audio or video of the sample image, to realize to the purpose explained with this page is drawn.
According to the above-mentioned course of work for drawing this interpreter, it is seen then that image retrieval technologies are to draw the function realization of this interpreter
Key technology point in process.And existing image retrieval technologies, because of the feature that amount of image information is big, often operand is big,
So time delay is longer.And for interaction class of electronic devices, handling duration is the important indicator for influencing Product Experience, so, how
The time delay for reducing image retrieval, becomes current urgent problem to be solved.
Summary of the invention
Applicant has found in the course of the study, existing image retrieval technologies, during retrieval, need to calculate to
Handle the distance between each word of each word and every sample image in image.And in fact, the word that every image includes
Quantity it is very big, meanwhile, for the accuracy for guaranteeing search result, general sample image quantity is also very big.So the process of retrieval
In, calculating distance between each word of image to be processed and each word of every sample image is the very big mistake of a calculation amount
It is slow to result in retrieval rate for journey, so that time delay is longer.
In view of this, the embodiment of the invention provides a kind of search method and devices, it is therefore intended that solve how to improve inspection
Suo Sudu, to reduce the problem of retrieving duration.
To achieve the above object, the embodiment of the present invention provides the following technical solutions:
A kind of search method, comprising:
Determine the matching word of object to be processed；
The distance parameter of preset candidate target is calculated, the distance parameter of any one candidate target is according to the matching word
Corresponding target range determines, wherein the corresponding target range of any one matching word are as follows: in the preset equivalent of the matching word
Target word at a distance from the matching word, the target word be include the word in the candidate target；The equivalent with it is described
The distance of matching word precalculates to obtain；
The candidate target that the distance parameter is met to preset condition, the search result as the object to be processed.
Above-mentioned method, optionally, the matching word of the determination object to be processed, comprising:
Extract the feature of the object to be processed；
The lexicographic tree constructed in advance is obtained, includes multilayer node in the lexicographic tree, it is any one in any one node layer
A node is a classification of next node layer；
The section for meeting preset condition at a distance from the feature is searched in next node layer of any one destination node
Point, until by node nearest with the characteristic distance in the last layer node, as the matching word of the object to be processed,
In, any one destination node is node nearest with the characteristic distance in same node layer.
Above-mentioned method, optionally, the process for constructing the lexicographic tree includes:
The feature of sample object is clustered, the dictionary containing multiple words is obtained, any one word is that cluster obtains one
A classification；
Using the word of the dictionary as the node of the last layer of the lexicographic tree；
For remaining any one layer of the lexicographic tree, is clustered according to next layer of the node to this layer, determine the layer
In node, any one node in this layer is the classification that cluster obtains, remaining described any one layer is the dictionary
In addition to any one layer of the last layer in tree.
Above-mentioned method optionally determines the process of the equivalent of the matching word, comprising:
According to the mark of the matching word, the equivalent of the matching word is found out in the adverbial word allusion quotation being pre-created；Institute
Stating adverbial word allusion quotation is the dictionary for recording the equivalent of each word and each word in dictionary, wherein the equivalent of any one word is
The preceding Q word being closer in the dictionary with the word.
Above-mentioned method optionally creates the process of the adverbial word allusion quotation, comprising:
Calculate the distance between every two word in the dictionary；
The equivalent of each word in the dictionary is obtained, the equivalent of any one word is the preceding Q nearest with word distance
A word, the Q are positive integer；
The corresponding equivalent and the distance between each word and equivalent for storing each word, each word, obtains
To the adverbial word allusion quotation.
Above-mentioned method optionally determines the distance parameter of the candidate target, comprising:
For any one candidate target, determine that the candidate is right according to first distance total value and/or second distance total value
The distance parameter of elephant；
Wherein, the first distance total value are as follows: the sum of the distance of the corresponding first object word of each matching word, arbitrarily
The corresponding first object word of one matching word are as follows: in the equivalent of the matching word, be included in the candidate target and with institute
Matching word is stated apart from nearest word；
The second distance total value are as follows: the sum of the distance of corresponding second target word of each matching word, any one
Corresponding second target word of the matching word are as follows: in the equivalent of the matching word, be included in the candidate target and with institute
State the close word of matching word distance second.
Above-mentioned method optionally determines the candidate target of the object to be processed, comprising:
Calculate inverse weight of each matching word in the object to be processed；
According to the glossarial index being pre-created, inverse weight of each matching word in sample object is obtained, wherein described
Each corresponding sample object of word and the word are in its corresponding sample pair in the preset dictionary that glossarial index storage precalculates
Inverse weight as in；
According to inverse weight of each matching word in the object to be processed and each matching word in correspondence
Sample object in inverse weight, determine the candidate target of the object to be processed.
Above-mentioned method optionally creates the process of the glossarial index, comprising: determines the corresponding sample of each word in dictionary
This object, the corresponding sample object of any one word are the sample object for including the word；
Calculate inverse weight of each word in the dictionary in corresponding sample object；
The corresponding word stored in the dictionary, the number of the corresponding sample object of institute's predicate and institute's predicate are corresponding
The inverse weight of sample object, obtains the glossarial index.
A kind of retrieval device, comprising:
Determining module, for determining the matching word of object to be processed；
Computing module, for calculating the distance parameter of preset candidate target, the distance parameter of any one candidate target
It is determined according to the corresponding target range of the matching word, wherein the corresponding target range of any one matching word are as follows: the matching word
Target word in preset equivalent at a distance from the matching word, the target word be include the word in the candidate target；Institute
Equivalent is stated to precalculate to obtain at a distance from the matching word；
Retrieval module, for the distance parameter to be met to the candidate target of preset condition, as the object to be processed
Search result.
A kind of retrieval facility, comprising: processor and memory, the memory is for storing application program, the processing
Device is for executing the application program, to realize above-mentioned search method.
A kind of computer readable storage medium is stored with instruction in the computer readable storage medium, when it is being calculated
When being run on machine, so that computer executes above-mentioned search method.
Search method provided in an embodiment of the present invention and device, determine the matching word of object to be processed, and calculate preset
Distance parameter is finally met the candidate target of preset condition by the distance parameter of candidate target, the retrieval as object to be processed
As a result.Wherein, the distance parameter of any one candidate target determines that the target range is according to the corresponding target range of matching word
For target word in the preset equivalent of the matching word at a distance from the matching word, target word is to include in the candidate target
Word.As it can be seen that search method described in the embodiment of the present application, it is no longer necessary to will be in the matching word and candidate target in object to be processed
Word calculate distance one by one, and only need to determine distance parameter according to target range.And equivalent is at a distance from matching word
(namely target word is at a distance from matching word) precalculates to obtain, i.e., the distance between corresponding word of matching word is being retrieved
Calculated completion before, need to only be obtained in retrieving target word in equivalent and the target word and matching word away from
From the distance parameter that can calculate candidate target.In conclusion method and device provided by the present application, during retrieval,
Without calculating object matching word to be processed at a distance from word a large amount of in dictionary, a large amount of calculating time is saved, to improve
The speed of retrieval.
Detailed description of the invention
In order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, to embodiment or will show below
There is attached drawing needed in technical description to be briefly described, it should be apparent that, the accompanying drawings in the following description is only this
The embodiment of invention for those of ordinary skill in the art without creative efforts, can also basis
The attached drawing of offer obtains other attached drawings.
Fig. 1 is a kind of flow chart of search method provided by the embodiments of the present application；
Fig. 2 is the schematic diagram of dictionary provided by the embodiments of the present application；
Fig. 3 is the flow chart of the matching word of determination provided by the embodiments of the present application image to be processed；
Fig. 4 is the flow chart of the candidate image of determination provided by the embodiments of the present application image to be processed；
Fig. 5 is the schematic diagram of glossarial index provided by the embodiments of the present application；
Fig. 6 is the flow chart of creation glossarial index provided by the embodiments of the present application；
Fig. 7 is the schematic diagram of adverbial word allusion quotation provided by the embodiments of the present application；
Fig. 8 is the flow chart of creation adverbial word allusion quotation provided by the embodiments of the present application；
Fig. 9 is the flow chart of the distance parameter provided by the embodiments of the present application for calculating preset candidate target；
Figure 10 is the flow chart of building lexicographic tree provided by the embodiments of the present application；
Figure 11 is another schematic diagram of adverbial word allusion quotation provided by the embodiments of the present application；
Figure 12 is the schematic diagram of lexicographic tree provided by the embodiments of the present application；
Figure 13 is another schematic diagram of lexicographic tree provided by the embodiments of the present application；
Distance parameter is met the candidate image of preset condition as figure to be processed to be provided by the embodiments of the present application by Figure 14
The flow chart of the search result of picture；
Figure 15 is search method architecture diagram provided by the embodiments of the present application；
Figure 16 is a kind of structural schematic diagram for retrieving device provided by the embodiments of the present application；
Figure 17 is a kind of structural schematic diagram of retrieval facility provided by the embodiments of the present application.
Specific embodiment
Search method and device disclosed in the embodiment of the present application, for retrieving the candidate for obtaining matching with object to be processed
Object.
The disclosed retrieval device of the embodiment of the present application can be applied on the electronic equipment with search function, may include
But it is not limited to early education class equipment.
Targeted object and candidate target to be processed, can include but is not limited to image, remove in embodiments herein
Except image, it can also be text, data (data that such as image zooming-out obtains, such as gray value data) etc., it below will be with
It is illustrated for image.
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete
Site preparation description, it is clear that described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.It is based on
Embodiment in the present invention, it is obtained by those of ordinary skill in the art without making creative efforts every other
Embodiment shall fall within the protection scope of the present invention.
Fig. 1 is a kind of search method disclosed in the embodiment of the present application, comprising the following steps:
S101: from preset dictionary, the matching word of object to be processed is determined.
In the present embodiment, preset dictionary is clustered by the feature to a large amount of sample object, to sample
The method that the feature of this object is clustered to obtain dictionary can refer to the prior art.Fig. 2 can be referred to by clustering obtained dictionary
Shown in schematic diagram, as shown in Fig. 2, dictionary is a two-dimensional array, the first of the two-dimensional array be classified as in dictionary in M word often
The number of a word, second is classified as the corresponding word of number.In the present embodiment, it should be noted that the number of word is the mark of word.
The matching word of object to be processed is word corresponding with the feature of object to be processed.In the present embodiment, number can be chosen
More, the sample object of A wide selection of colours and designs is measured, a large amount of differences of the dictionary for obtaining cluster including the matching word of object to be processed
Word, it is ensured that the matching word of object to be processed can be found out in the dictionary.Specifically determine the matching word of object to be processed
Process can refer to the prior art, also may refer to process shown in Fig. 3.
S102: according to matching word, the candidate target of object to be processed is obtained.
In the present embodiment, candidate target be can be in a large amount of sample object, higher with object similarity degree to be processed
Sample object determines that the process of the corresponding candidate target of object to be processed can be with reference to the prior art, can also be with reference to shown in Fig. 4
Process.Alternatively, candidate target can also be regard whole sample objects as.
S103: the word for including in candidate target is obtained.
In the present embodiment, the word that includes in candidate target be candidate target feature in dictionary corresponding word.It can be according to
The word for including in candidate target is obtained according to the glossarial index being pre-created.The schematic diagram of glossarial index may refer to Fig. 5.As shown in figure 5,
Glossarial index includes word all in dictionary, and each word has a corresponding two-dimensional array, the first row of the two-dimensional array
The word is classified as in the inverse weight of the sample object for the number of sample object, second.Glossarial index has recorded each sample object
The equivalent of the number of number and each sample object, the corresponding word of the number of any one sample object are the sample object
In include word.Mapping relations are designed between the corresponding word of the number of any one sample object, are closed by the mapping
System can find out the corresponding word of sample object.
The specific process for obtaining the word for including in candidate target are as follows: in glossarial index, find out the number with candidate target
Identical picture number, and obtained corresponding with the target designation using the picture number as target designation by mapping relations
Whole words, and using whole word corresponding with the target designation as the word for including in the candidate image.Wherein, word rope is created
The process drawn can refer to process shown in fig. 6.
In the present embodiment, determining that the word for including in candidate target is also possible that is each spy of candidate target by dictionary
Sign matches corresponding word, and specific matching process is identical as the process of matching word for determining object to be processed, and details are not described herein again.
S104: the preset equivalent of matching word is obtained.
In the present embodiment, the equivalent of matching word is Q word before being closer in dictionary with the matching word, wherein any
The distance between one corresponding word of matching word is the distance precalculated stored in the adverbial word allusion quotation being pre-created.Adverbial word
The schematic diagram of allusion quotation may refer to Fig. 7, as shown in fig. 7, the adverbial word allusion quotation includes an one-dimensional array, each element of the array
For each word for carrying number in dictionary.Each word in array corresponds to a two-dimensional array, the first row of the two-dimensional array
For the preceding Q word nearest with word distance, second is classified as the word at a distance from each word in preceding Q word.Wherein, adverbial word is created
The detailed process of allusion quotation may refer to process shown in Fig. 8.
In the present embodiment, the process of the equivalent of matching word is obtained specifically: from a dimension of the adverbial word allusion quotation being pre-created
Word identical with the number of matching word is found out in group, each word of first row is i.e. in the corresponding two-dimensional array of the word found
For the equivalent of matching word.
S105: according to the word for including in the equivalent and candidate target of matching word, the preset candidate target is calculated
Distance parameter.
In the present embodiment, the distance parameter of any one candidate target is determined according to the corresponding target range of matching word,
With the corresponding target range of word are as follows: the target word in the preset equivalent of the matching word at a distance from the matching word, meanwhile, target
Word be include the word in the candidate target.The process for calculating the distance parameter of preset candidate target can be with reference to shown in Fig. 9
Process.
Distance parameter: being met the candidate target of preset condition by S106, the search result as object to be processed.
In the present embodiment, preset condition can be threshold condition, and different threshold values can be set to different distance parameters
Condition, threshold condition can be set in conjunction with actual conditions, for example, if sample image is a drawing sheet, figure to be processed
Cover as being the drawing sheet, then threshold condition can be set to cover threshold value, which is the image for cover type
The threshold value of setting.Finally, the candidate target of threshold condition will be met as the search result of object to be processed.
Existing image retrieval technologies, during retrieval, need to calculate object to be processed matching word and each sample
The distance between each word for including in this object, slow so as to cause retrieval rate, time-consuming.Side provided in this embodiment
Method first determines candidate target from a large amount of sample object, and further determines that the distance parameter of candidate target, finally obtains inspection
The result of rope.Distance parameter is to be calculated by target range, and target range is the matching word and target word precalculated
The distance between, which can directly get from adverbial word allusion quotation.So need to only be obtained from adverbial word allusion quotation in retrieving
The target range is taken, the distance parameter of candidate target can be calculated, without calculating the matching word of object to be processed and each
The distance for the word for including in sample object.In conclusion method provided in this embodiment, has saved a large amount of retrieval time, from
And improve the speed of retrieval.
Below will using " image " as Fig. 1 in " object " specific implementation, in Fig. 1 each step carry out it is detailed
Explanation.
Fig. 3 determines a kind of embodiment of the matching word of image to be processed for the S101 in Fig. 1, specifically includes:
S301: the feature of image to be processed is extracted.
In the present embodiment, the feature of image to be processed can be extracted using a kind of method arbitrary in SIFT, URF and ORB,
Without limitation to the method for extracting feature, the method for specifically extracting feature can refer to the prior art to the present embodiment, herein no longer
It repeats.
It should be noted that the feature that image zooming-out to be processed arrives is a large amount of feature, S302~S304 performed below
To determine the process of the matching word of this feature for any one feature.
S302: the lexicographic tree constructed in advance is obtained.
In the present embodiment, lexicographic tree is the tree structure obtained to the word progress multistratum classification in dictionary constructed in advance
Dictionary, a node in the lexicographic tree are a classification of next node layer.The process of building lexicographic tree is referred to Figure 10
Shown in flow chart.
S303: at a distance from each node for calculating the feature of image to be processed and the first layer of lexicographic tree, selected distance is most
Close node is as first object node.
In the present embodiment, if feature is closer with nodal distance, illustrate that this feature and the matching degree of the node are higher.It needs
It is noted that the case where for special circumstances, i.e. only one node of lexicographic tree first layer, this node of first layer is made
S302 is skipped then after executing S301 for first object node, directly execution S303.
S304: will meet at a distance from feature in next node layer of first object node the node of preset condition as
Second destination node.
In the present embodiment, next node layer of any one destination node is the branch node of the destination node, arbitrarily
The node for meeting preset condition is node nearest with object distance to be processed in same node layer.In first object node
Meet at a distance from feature in next node layer preset condition node be in next node layer of first object node with
This feature is apart from nearest node.
S305: the node of preset condition will be met in next node layer of the second destination node as third target section
Point, and so on, until finding the last one destination node in the node of the last layer.
In the present embodiment, by successively searching each layer of destination node, until being found in the node of the last layer
The destination node of the last layer, the destination node of the last layer are the last one destination node.
S306: using the destination node of the last layer as the matching word of image to be processed.
In the present embodiment, by layer-by-layer matched method, by first layer to last one layer of sequence in lexicographic tree, successively
It obtains for the characteristic matching of image to be processed in each layer apart from nearest word, finally by the last layer of lexicographic tree and the spy
Levy matching word of the destination node as this feature apart from nearest word namely the last layer.
In the prior art, feature is successively calculated at a distance from each word in dictionary, is finally selected apart from nearest word work
The matching word being characterized, due in feature and dictionary word sum it is huge, and cause matching time-consuming.It is provided in this embodiment
Method is chosen with the nearest node of characteristic distance in each layer of lexicographic tree as target using the lexicographic tree pre-established
Node, until selecting in the last layer with the specific word apart from nearest node as the matched word of this feature.By above-mentioned
Method, only need the calculating of at most (L-1) * B number that matching word of this feature in dictionary can be obtained, wherein L and B are respectively
The number of plies of lexicographic tree and the branch amount of destination node.And the number of plies of lexicographic tree and the branch amount of destination node are less, so
Matched efficiency can be improved by above-mentioned matching process, to improve the speed of retrieval.
Fig. 4 be Fig. 1 in S102 determine image to be processed candidate image a kind of embodiment, can specifically include with
Lower step:
S401: inverse weight of each matching word in image to be processed is calculated.
In the present embodiment, inverse weight of any one matching word in image to be processed can be calculated according to formula (1) and be obtained
?.
Wherein, pre_tfidf_word (i) indicates inverse weight of the matching word in image to be processed, and i indicates the number of word,
Npre_feature indicates that the sum of the feature of image to be processed, npre_word (i) indicate what the matching word that number is i occurred
Number.
S402: according to the glossarial index being pre-created, inverse weight of each matching word in sample image is obtained.
In the present embodiment, as in the foregoing embodiment, the corresponding sample of each word in the dictionary that glossarial index storage precalculates
The inverse weight of this image and the word in its corresponding sample image, so can be looked for by the number of matching word in glossarial index
The matching word out, and obtain inverse weight of the matching word in sample image.
S403: according to inverse weight of each matching word in image to be processed and each matching word in corresponding sample graph
Inverse weight as in, calculate the weight of sample image with.
In the present embodiment, the weight of sample image and the matching score for indicating sample image and image to be processed, score are got over
Height indicates that a possibility that image to be processed is with the sample image successful match is bigger.The matching score can be counted according to formula (2)
It calculates and obtains.
Wherein, iw (i) (j) indicates inverse weight of the word i in image j, and identical symbol can refer to formula (1) for remaining
Symbolic interpretation in formula (1), details are not described herein again.
S404: according to each sample image weight and, obtain the candidate image of image to be processed.
In the present embodiment, optionally, descending arrangement is carried out to the score value for the sum_w (j) that formula (2) is calculated, will be arranged
The preceding n sample image of sequence is as candidate image.
Method provided in this embodiment, preset glossarial index are pre-calculated the matching word of image to be processed in each sample
Inverse weight in picture is not necessarily to so inverse weight of the matching word in each samples pictures obtains directly from glossarial index
It is calculated.Meanwhile the present embodiment is using weight in each sample image and maximum preceding n sample image as image to be processed
Candidate image, it is not necessary that a whole sample images as candidate, are improved the speed of retrieval, and the power by calculating sample image
The candidate image selected again and the matching degree of image to be processed are bigger, and accuracy is higher.In conclusion the present embodiment provides
Method, the speed of retrieval can not only be improved and improve the accuracy of retrieval.
Fig. 6 is the process for creating glossarial index, can specifically include following steps:
S601: the corresponding sample image of word in dictionary is determined.
It include a large amount of word in each sample image, so the same word appears in different samples in the present embodiment
In this image, the corresponding sample image of any one word is the sample image for including the word.
S602: inverse weight of the word in dictionary in corresponding sample image is determined.
In the present embodiment, inverse weight of the word in its corresponding sample image can be calculated according to formula (3) and be obtained.
Wherein, iw (i) (j) indicates inverse weight of the word i of sample in sample image j, and n_eature (j) indicates image j
The sum of appearance, n_ (i) (j) indicate the number that matching word i occurs in image j.
S603: the number of word, the corresponding sample image of the word in corresponding storage dictionary and the word are in corresponding sample graph
Inverse weight as in creates glossarial index.
In the present embodiment, the number of the word in dictionary is set as first row, second is classified as word corresponding with number.Because every
A sample image has unique number, thus number can representative sample image, by the number of the corresponding each sample image of word
As third column data, using inverse weight of the word in each sample image as the 4th column data, and by third column data with
4th column data is spliced to obtain the corresponding two-dimensional array of the word.It sets between the corresponding two-dimensional array of the word
Mapping relations make that two-dimensional array corresponding with the word can be found by the number of the word or the word.Wherein, mapping relations are set
Surely the prior art can be referred to.
The word of different numbers each in dictionary is adopted with the aforedescribed process, the corresponding two-dimemsional number of each word can be obtained
Group.By in dictionary each word and the corresponding two-dimensional array of each word be recorded in data file, and by the record
Document is as glossarial index.
Method provided in this embodiment, by the corresponding sample image of each word of determination and the word in its corresponding sample
Inverse weight in this image generates glossarial index.Since glossarial index generates before retrieval, so during retrieval,
The corresponding samples pictures of matching word that image to be processed is directly obtained by glossarial index, and inverse weight in samples pictures is
Can, without being calculated, reduce the calculation amount of retrieval, to improve the speed of retrieval.
Fig. 8 is a kind of embodiment for creating adverbial word allusion quotation, can specifically include following step:
S801: the distance between every two word in dictionary is calculated.
In the present embodiment, the first word can be successively calculated by two identical dictionaries, the first dictionary and the second dictionary
Each word in allusion quotation is at a distance from the word in the second dictionary in addition to itself, thus obtain in dictionary between every two word away from
From.It can also be and only use a dictionary, successively calculate each word at a distance from remaining word in dictionary in addition to itself, thus
Obtain the distance between every two word in dictionary.
It should be noted that the word different for two, such as word A and word B, when be computed for the first time word A to word B it
Between distance after, second can not calculate the distance of word B to word A, directly acquire between the A to word B that first time is calculated
Distance value.
S802: the equivalent of each word in dictionary is obtained.
In the present embodiment, by the available each word of S801 and dictionary in addition to itself the distance between remaining word,
The equivalent of any one word is preceding Q word nearest with word distance in dictionary.Wherein, Q is positive integer, and the size of Q can be with
Sets itself.
S803: corresponding to store the distance between each word, the equivalent of each word and the corresponding word of each word, creation
Adverbial word allusion quotation.
In the present embodiment, described in embodiment as the aforementioned, each word is the word for carrying number, is to number the word for being 4
Example will number the first column data of conduct of the equivalent for the word for being 4, will number the distance between corresponding word of word for being 4 and make
For the second column data, and the first column data and the second column data are spliced into a two-dimensional array.Setting number is 4 words and should
Mapping relations between two-dimensional array make through number to be that 4 words can find two-dimensional array corresponding with the word.To every in dictionary
The word of a different numbers all uses the above-mentioned method as numbered the word for 4, and the corresponding two-dimensional array of each word can be obtained.It will
Each word and the corresponding two-dimensional array of each word in dictionary are recorded in data file, and by the recording documents
As adverbial word allusion quotation.
Optionally, in adverbial word allusion quotation, the word of the carrying number of record can be replaced with to the number of the word, because of the number of word
It is correspondingly, so the number of word can indicate the word with word.The number of word indicates the signal of the adverbial word allusion quotation of the word
Figure can refer to Figure 11.
Method provided in this embodiment calculates the distance between every two word in dictionary, for each word find with the word away from
Mapping relations are set from similar each word, and for the corresponding word of each word, finally create adverbial word allusion quotation.Make the process in retrieval
In, without being calculated, that is, it can determine the distance between equivalent and the corresponding word of each word of each word, to mention
The high retrieval rate to image to be processed.
Fig. 9 calculates a kind of embodiment of distance parameter of preset candidate image for the S105 in Fig. 1, specifically includes following
Step:
S901: according to the equivalent of matching word and the word of candidate image, the corresponding first object word of matching word is found out
The distance of distance and the second target word.
In the present embodiment, by preset adverbial word allusion quotation, the equivalent of any one matching word can be determined, in matching word
In equivalent, the distance between matching word and each equivalent are different from, according to the distance between corresponding word of matching word
Sequence from small to large finds out word identical with the number of the equivalent, and first will found in the word of candidate image
The first object word as the matching word of the word of a identical candidate image of number with equivalent, second found with
Second target word of the word of the identical candidate image of number of equivalent as the matching word, while by matching word and first object
Distance of the distance of word as first object word, by distance of the matching word at a distance from the second target word as the second target word.
It will include making with the matching word apart from nearest word in the equivalent of matching word and the word of candidate image
For first object word, the word close with the matching word distance second is as the second target word.
S902: according to the distance of first object word and the distance of the second target word, obtain first distance total value and second away from
From total value.
In the present embodiment, optionally, to the distance of the corresponding first object word of each matching word of image to be processed into
Row superposition summation, obtains first distance total value, similarly, to corresponding second target word of each matching word of image to be processed
Distance is overlapped summation, obtains second distance total value.
S903: according to first distance total value and second distance total value, the distance parameter of the candidate image is obtained.
, optionally, can be using first distance total value as the distance parameter of the candidate image in the present embodiment, it can also be by one
Distance parameter of the difference as the candidate image apart from total value and second distance total value, can also using second distance total value as
The distance parameter of the candidate image.
It should be noted that in the present embodiment, the distance parameter of each candidate image be by above-mentioned S901~
S903 whole process is calculated.
In the present embodiment, the distance parameter of candidate image is calculated by first distance total value and second distance total value, first
It is the summation of the distance of the corresponding first object word of each matching word apart from total value, second distance total value is each matching word
The summation of the distance of corresponding second target word, and the distance of any one first object word and any one second target word
Distance can directly be obtained from adverbial word allusion quotation, without being calculated, reduce the amount of calculation of retrieval, to improve
The speed of retrieval.
Figure 10 is a kind of embodiment of the S302 building lexicographic tree in Fig. 3, can specifically include following steps:
S1001: clustering the feature of sample image, obtains the dictionary containing multiple words.
In the present embodiment, sample image is pre-prepd training image, extracts the feature of every sample image, and to mentioning
The feature taken is clustered, the dictionary containing multiple words of most Zhongdao, any one word is to obtain a classification by cluster, i.e., often
A word is the information word for characterizing a category feature, and the corresponding word of the high feature of same or similar degree is identical.Wherein, image is extracted
Feature and the method clustered to feature can refer to the prior art.
S1002: the number of plies of lexicographic tree and the branch amount of each node are preset.
In the present embodiment, the branch amount of the number of plies of lexicographic tree and each node can by word in dictionary total quantity really
It is fixed.For example, the sum of word is M, number of plies L, when the branch amount of each node is B, optionally, L and B can be by formula M=BLReally
It is fixed.For M=B1L1, M=B2L2, the case where wherein B1 ≠ B2, L1 ≠ L2, one can be selected in the two i.e. in conjunction with actual demand
It can.It should be noted that the relational expression of M and L and B can also be determined by M=B*L, the present embodiment to the relational expression of M and L and B not
It limits.
Optionally, in addition to the total quantity according to word in dictionary sets the number of plies of lexicographic tree and the branch amount of each node,
It can also be taking human as the branch amount for the number of plies and each node for rule of thumb directly setting lexicographic tree.
S1003: the number of plies and each node branch number according to lexicographic tree classify to the word in dictionary, and will divide
Node of the word as lexicographic tree the last layer after class.
In the present embodiment, due to containing a large amount of different words in dictionary, the distance between some difference words have a long way to go,
Therefore need to carry out the word in dictionary preliminary classification, it marks off and belongs to same category of word, which is each
The word being closer between word.The quantity of classification and the number of plies of preset lexicographic tree and each node branch that classification obtains
Number is related, for example, if in dictionary word sum M=8, according to M=BL, then 8=23.In order to guarantee the number of plies of the lexicographic tree created
It is 3 layers, the branch amount of each node is 2, then needs every 2 closely located words in 8 words being divided into one kind.If in dictionary
Sum M=9 of word, according to 9=33, in order to guarantee that the number of plies of the lexicographic tree of creation is 3 layers, the branch amount of each node is 3, then
Every 3 closely located words in 9 words need to be divided into one kind.Finally using sorted each word as lexicographic tree last
The node of layer.
S1004: the node that any one layer of remaining of lexicographic tree carries out cluster determination according to next layer of the node to this layer.
In the present embodiment, remaining any one layer is remaining any one layer in lexicographic tree in addition to the last layer.It chooses last
Closely located node (i.e. what classification obtained in S1003 belongs to same category of node) is connected to same section in one layer of node
Point, and using the same node as a node of layer second from the bottom, wherein the number of node similar in selected distance and default
Each node branch amount it is identical, finally can be obtained layer second from the bottom each node and layer second from the bottom it is each
Node and most one layer of connection relationship.
Similarly, classify to the node of layer second from the bottom, closely located node is connected to same node, and this is same
Each node of layer third from the bottom and third from the bottom finally can be obtained in a node of one node as layer third from the bottom
Each node of layer and the connection relationship of layer second from the bottom.Analogize in this, the first layer of lexicographic tree finally can be obtained, to obtain
Lexicographic tree, the lexicographic tree created may refer to Figure 12.
Optionally, the method for creating lexicographic tree may also is that point of the number of plies and each node without limiting lexicographic tree
Branch number, clusters after obtaining the dictionary containing multiple words the feature of candidate image, is directly classified to the word in dictionary,
And using sorted word as the node of the last layer of lexicographic tree, the closely located node of selection the last layer is connected to same
Node, and using the same node as a node of layer second from the bottom, so that it is determined that each section of lexicographic tree layer second from the bottom
The connection relationship of each node and the last layer of point and layer second from the bottom, to the processing method of layer second from the bottom with last
The processing method of layer is the same, so that layer third from the bottom is obtained, and so on, the first layer of lexicographic tree finally can be obtained, thus
To lexicographic tree, the lexicographic tree created may refer to Figure 13.
Method provided in this embodiment obtains the dictionary containing multiple words by clustering to the feature of candidate image,
And lexicographic tree is created according to the word of dictionary, the lexicographic tree of creation carries out multistratum classification to each word of dictionary, makes the mistake in retrieval
Cheng Zhong quickly can be that the characteristic matching of image to be processed, apart from nearest word, is improved to this feature by lexicographic tree
The speed of retrieval.
Method provided by the embodiments of the present application, when sample image is a drawing sheet, image to be processed is the drawing sheet
Cover and drawing this in when drawing this page, distance parameter is met into the candidate image of preset condition as image to be processed in Fig. 1
The specific process as shown in figure 14 of search result:
S1401: it is initialized.
In the present embodiment, cover threshold condition can be preset in initial phase and draw this threshold condition.
S1402: the first distance total value, second distance total value and first distance total value of each candidate image and the are obtained
Two differences apart from total value.
S1403: judging that image to be processed still draws this page for cover, hold S1404~S1407 if it is cover, if it is
It draws this page and executes S1408~S1411.
S1404: the difference of the maximum candidate image of first distance total value or first distance total value and second distance total value is obtained
It is worth the smallest candidate image as best candidate image；
In the present embodiment, the first distance total value of candidate image is bigger or the difference of first distance total value and second distance total value
It is worth smaller, illustrates that candidate image and the matching degree of image to be processed are higher.
S1405: judge the first distance total value or first distance total value and second distance total value of the best candidate image
Whether difference is greater than cover threshold value, if so, executing S1406, export the best candidate image as a result, if not, execute
S1407 does not export result.
In the present embodiment, the image richness of cover is similar, therefore can be met using a set of cover threshold value and be got correctly
Search result demand.
S1408: the difference of the maximum candidate image of first distance total value or first distance total value and second distance total value is obtained
It is worth the smallest candidate image as best candidate image.
S1409: judge the first distance total value or first distance total value and second distance total value of the best candidate image
Whether difference less than first draws this threshold value, if so, execute S1406, export the best candidate image as a result, if not, holding
Row S1410.
In the present embodiment, the identification of this page is drawn in this for drawing, there are biggish differences each other since each page draws this page
It is different, it is a set of draw this threshold value identification can no longer meet requirement, need to carry out the judgement of the second subthreshold.
S1410: the absolute value abs_id of the number difference of the best candidate image and the last search result is calculated
Whether less than 2, if so, S1411 is executed, if not, executing S1407.
In the present embodiment, abs_id < 2 indicates that the content pages currently identified are the front and backs page two of last time recognition result.
S1411: whether the difference of the first distance total value that judges the best candidate image and second distance total value is greater than the
Two draw this threshold value, if so, S1406 is executed, if not, executing S1407.
In the present embodiment, continue the first distance total value for judging the best candidate image and second distance if abs_id < 2
Whether the difference of total value, which is greater than second, is drawn this threshold value, wherein second draws this threshold value than first to draw this threshold value small.
Method provided in this embodiment, for sample image be drawing this, image to be processed be this risk of drawing or
The case where this drawing paint also sets different threshold conditions, to improve the accuracy of retrieval.
In conclusion method provided by the embodiments of the present application can be realized by search method architecture diagram disclosed in Figure 15.
The architecture diagram includes training set bottom library, rough matching model and accurate Matching Model.
The training set bottom library includes the dictionary being pre-created, lexicographic tree, adverbial word allusion quotation and glossarial index.
Wherein dictionary, lexicographic tree, adverbial word allusion quotation, the creation process of glossarial index can refer to above-mentioned each embodiment, herein not
It repeats again.
The workflow of rough matching model are as follows:
If receiving image to be processed, the feature of image to be processed is obtained.
The corresponding matching word of feature for finding image to be processed using lexicographic tree.
The equivalent of matching word is found using adverbial word allusion quotation.
Matching word is calculated in the inverse weight in image to be processed.
Matching word is obtained in the inverse weight of each sample image using glossarial index.
It is obtained in conjunction with inverse weight of the matching word in image to be processed and the inverse weight calculation in each sample image each
The weight of sample image and.
According to the sequence of weight and descending, using preceding n sample image as candidate image.
The workflow of accurate Matching Model are as follows:
Using adverbial word allusion quotation, successively obtain the first distance total value of each candidate image, second distance total value and first away from
Difference from total value and second distance total value.
The most preceding candidate image or by first distance total value and the of sorting is obtained according to the sequence of first distance total value ascending order
Two sequences apart from total value difference descending obtains most preceding candidate image of sorting.
Search result of the candidate image that will acquire as image to be processed.
Method provided in this embodiment builds dictionary, lexicographic tree, adverbial word allusion quotation and word rope by training set bottom library in advance
Draw, make during being retrieved using rough matching model, rough matching model can be directly from the library of training set bottom directly
Corresponding data are obtained, rough matching model is exported without largely being calculated, while using accurate Matching Model
Result further screened, finally obtained the matching result of image to be processed.In conclusion provided in this embodiment
The speed of retrieval not only can be improved in method, but also improves the accuracy of retrieval.
Corresponding with the method for Fig. 1, the embodiment of the invention also provides retrieval devices, for the specific reality to Fig. 1 method
Existing, structural schematic diagram is as shown in figure 16, specifically includes:
Determining module 1601, for determining the matching word of object to be processed.
Computing module 1602, for calculating the distance parameter of preset candidate target, the distance of any one candidate target
Parameter is determined according to the corresponding target range of matching word, wherein the corresponding target range of any one matching word are as follows: the matching word
Target word in preset equivalent at a distance from the matching word, target word be include the word in the candidate target；Equivalent
With precalculate to obtain at a distance from matching word.
Retrieval module 1603, the inspection for distance parameter to be met to the candidate target of preset condition, as object to be processed
Hitch fruit.
Above-mentioned retrieval device, further includes:
Lexicographic tree module 1604 is constructed, is clustered for the feature to sample object, obtains the dictionary containing multiple words,
Any one word is that cluster obtains a classification；Using the word of dictionary as the node of the last layer of lexicographic tree；For lexicographic tree
Remaining any one layer clusters according to next layer of the node to this layer, determines the node in this layer, any one in this layer
A node is the classification that cluster obtains, remaining any one layer is any one layer in lexicographic tree in addition to the last layer.
Secondary lexicon module 1605 is generated, for calculating the distance between every two word in dictionary；Obtain each word in dictionary
Equivalent, the equivalent of any one word is the preceding Q word nearest with word distance, and Q is positive integer；Corresponding storage is each
The equivalent and the distance between each word and equivalent of word, each word, obtain adverbial word allusion quotation.
Glossarial index module 1606 is generated, for determining the corresponding sample object of each word in dictionary, any one word is corresponding
Sample object be the sample object for including the word；Calculate inverse weight of each word in dictionary in corresponding sample object；
The number and word of word, the corresponding sample object of word in corresponding storage dictionary are obtained in the inverse weight of corresponding sample object
Glossarial index.
Wherein it is determined that module 1601, determines the specific implementation of the matching word of object to be processed are as follows: it is to be processed right to extract
The feature of elephant；The lexicographic tree constructed in advance is obtained, includes multilayer node in lexicographic tree, any one section in any one node layer
Point is a classification of next node layer；It searches in next node layer of any one destination node and meets at a distance from feature
The node of preset condition, until the matching by node nearest with characteristic distance in the last layer node, as object to be processed
Word, wherein any one destination node is node nearest with characteristic distance in same node layer.
Wherein, computing module 1602 determine the specific implementation of the equivalent of matching word are as follows: the mark according to matching word
Know, the equivalent of matching word is found out in the adverbial word allusion quotation being pre-created；Adverbial word allusion quotation is for each word in record dictionary and each
The dictionary of the equivalent of word, wherein the equivalent of any one word is the preceding Q word being closer in dictionary with the word.
Wherein, computing module 1602 determine the specific implementation of the distance parameter of candidate target are as follows: for any one
Candidate target determines the distance parameter of candidate target according to first distance total value and/or second distance total value；Wherein, first away from
From total value are as follows: the sum of the distance of the corresponding first object word of each matching word, the corresponding first object word of any one matching word
Are as follows: in the equivalent of matching word, including in candidate target and with matching word apart from nearest word；Second distance total value are as follows: each
The sum of the distance of corresponding second target word of a matching word, corresponding second target word of any one matching word are as follows: matching word
In equivalent, including in candidate target and the word close with matching word distance second.
Wherein, computing module 1602 determine the specific implementation of the candidate target of object to be processed are as follows: calculate each
Inverse weight with word in object to be processed；According to the glossarial index being pre-created, each matching word is obtained in sample object
Inverse weight, wherein each corresponding sample object of word and the word are at it in the preset dictionary that glossarial index storage precalculates
Inverse weight in corresponding sample object；Exist according to inverse weight of each matching word in object to be processed and each matching word
Inverse weight in corresponding sample object, determines the candidate target of object to be processed.
Retrieval device provided in an embodiment of the present invention, determines the matching word of object to be processed, and it is right to calculate preset candidate
Distance parameter is finally met the candidate target of preset condition by the distance parameter of elephant, the search result as object to be processed.Its
In, the distance parameter of any one candidate target determines that the target range is the matching according to the corresponding target range of matching word
Target word in the preset equivalent of word at a distance from the matching word, target word be include the word in the candidate target.As it can be seen that
The retrieval device of the embodiment of the present application, it is no longer necessary to calculate the word in the matching word and candidate target in object to be processed one by one
Distance, and only need to determine distance parameter according to target range.And equivalent at a distance from matching word (namely target word with
The distance of matching word) it precalculates to obtain, i.e., the distance between corresponding word of matching word has been computed before being retrieved
It completes, the target word and the target word that need to be only obtained in equivalent in retrieving can calculate time at a distance from matching word
Select the distance parameter of object.In conclusion device provided by the present application, during retrieval, without calculating object to be processed
Matching word has saved a large amount of calculating time, to improve the speed of retrieval at a distance from word a large amount of in dictionary.
The embodiment of the invention also provides retrieval facilities, and structural schematic diagram is as shown in figure 17, including 1701 He of processor
Memory 1702, memory is for storing program, and processor is for running program, and to realize, pose is determined in the embodiment of the present invention
Method.
In this application, the terms "include", "comprise" or any other variant thereof is intended to cover non-exclusive inclusion,
So that the process, method, article or equipment for including a series of elements not only includes those elements, but also including not having
The other element being expressly recited, or further include for elements inherent to such a process, method, article, or device.Do not having
There is the element limited in the case where more limiting by sentence "including a ...", it is not excluded that in the mistake including the element
There is also other identical elements in journey, method, article or equipment.
The foregoing description of the disclosed embodiments enables those skilled in the art to implement or use the present invention.
Various modifications to these embodiments will be readily apparent to those skilled in the art, as defined herein
General Principle can be realized in other embodiments without departing from the spirit or scope of the present invention.Therefore, of the invention
It is not intended to be limited to the embodiments shown herein, and is to fit to and the principles and novel features disclosed herein phase one
The widest scope of cause.
Claims (11)
1. a kind of search method characterized by comprising
Determine the matching word of object to be processed；
The distance parameter of preset candidate target is calculated, the distance parameter of any one candidate target is corresponding according to the matching word
Target range determine, wherein the corresponding target range of any one matching word are as follows: the mesh in the preset equivalent of the matching word
Mark word at a distance from the matching word, the target word be include the word in the candidate target；The equivalent and the matching
The distance of word precalculates to obtain；
The candidate target that the distance parameter is met to preset condition, the search result as the object to be processed.
2. the method according to claim 1, wherein the matching word of determination object to be processed, comprising:
Extract the feature of the object to be processed；
The lexicographic tree constructed in advance is obtained, includes multilayer node in the lexicographic tree, any one section in any one node layer
Point is a classification of next node layer；
The node for meeting preset condition at a distance from the feature is searched in next node layer of any one destination node, directly
To the matching word by node nearest with the characteristic distance in the last layer node, as the object to be processed, wherein appoint
Destination node of anticipating is node nearest with the characteristic distance in same node layer.
3. according to the method described in claim 2, it is characterized in that, the process for constructing the lexicographic tree includes:
The feature of sample object is clustered, the dictionary containing multiple words is obtained, any one word is that cluster obtains a class
Not；
Using the word of the dictionary as the node of the last layer of the lexicographic tree；
For remaining any one layer of the lexicographic tree, is clustered, determined in this layer according to next layer of the node to this layer
Node, any one node in this layer are the classification that cluster obtains, remaining described any one layer is in the lexicographic tree
In addition to any one layer of the last layer.
4. the method according to claim 1, wherein determining the process of the equivalent of the matching word, comprising:
According to the mark of the matching word, the equivalent of the matching word is found out in the adverbial word allusion quotation being pre-created；
The adverbial word allusion quotation is the dictionary for recording the equivalent of each word and each word in dictionary, wherein pair of any one word
Answering word is the preceding Q word being closer in the dictionary with the word.
5. according to the method described in claim 4, it is characterized in that, creating the process of the adverbial word allusion quotation, comprising:
Calculate the distance between every two word in the dictionary；
The equivalent of each word in the dictionary is obtained, the equivalent of any one word is the preceding Q word nearest with word distance,
The Q is positive integer；
The corresponding equivalent and the distance between each word and equivalent for storing each word, each word, obtains institute
State adverbial word allusion quotation.
6. the method according to claim 1, wherein determining the distance parameter of the candidate target, comprising:
For any one candidate target, the candidate target is determined according to first distance total value and/or second distance total value
Distance parameter；
Wherein, the first distance total value are as follows: the sum of the distance of the corresponding first object word of each matching word, any one
The corresponding first object word of matching word are as follows: in the equivalent of the matching word, be included in the candidate target and with described
With word apart from nearest word；
The second distance total value are as follows: the sum of the distance of corresponding second target word of each matching word, it is any one of
Corresponding second target word of matching word are as follows: in the equivalent of the matching word, be included in the candidate target and with described
With the close word of word distance second.
7. the method according to claim 1, wherein determining the candidate target of the object to be processed, comprising:
Calculate inverse weight of each matching word in the object to be processed；
According to the glossarial index being pre-created, inverse weight of each matching word in sample object is obtained, wherein institute's predicate rope
Draw in the preset dictionary that precalculates of storage each corresponding sample object of word and the word in its corresponding sample object
Inverse weight；
According to inverse weight of each matching word in the object to be processed and each matching word in corresponding sample
Inverse weight in this object determines the candidate target of the object to be processed.
8. the method according to the description of claim 7 is characterized in that creating the process of the glossarial index, comprising:
Determine the corresponding sample object of each word in dictionary, the corresponding sample object of any one word is the sample pair for including the word
As；
Calculate inverse weight of each word in the dictionary in corresponding sample object；
The corresponding word stored in the dictionary, the number of the corresponding sample object of institute's predicate and institute's predicate are in corresponding sample
The inverse weight of object, obtains the glossarial index.
9. a kind of retrieval device characterized by comprising
Determining module, for determining the matching word of object to be processed；
Computing module, for calculating the distance parameter of preset candidate target, the distance parameter foundation of any one candidate target
The corresponding target range of the matching word determines, wherein the corresponding target range of any one matching word are as follows: the matching word is default
Equivalent in target word at a distance from the matching word, the target word be include the word in the candidate target；It is described right
Word is answered to precalculate to obtain at a distance from the matching word；
Retrieval module, the inspection for the distance parameter to be met to the candidate target of preset condition, as the object to be processed
Hitch fruit.
10. a kind of retrieval facility characterized by comprising
Processor and memory, the memory are used to execute the application program for storing application program, the processor,
To realize the described in any item search methods of claim 1-8.
11. a kind of computer readable storage medium, which is characterized in that instruction is stored in the computer readable storage medium,
When run on a computer, so that computer perform claim requires the described in any item search methods of 1-8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910567616.XA CN110263198A (en) | 2019-06-27 | 2019-06-27 | A kind of search method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910567616.XA CN110263198A (en) | 2019-06-27 | 2019-06-27 | A kind of search method and device |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110263198A true CN110263198A (en) | 2019-09-20 |
Family
ID=67922333
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910567616.XA Pending CN110263198A (en) | 2019-06-27 | 2019-06-27 | A kind of search method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110263198A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111191067A (en) * | 2019-12-25 | 2020-05-22 | 深圳市优必选科技股份有限公司 | Picture book identification method, terminal device and computer readable storage medium |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20090303511A1 (en) * | 2005-12-19 | 2009-12-10 | Xerox Corproration | Database system for processing image data for consistent color printing |
| CN103020111A (en) * | 2012-10-29 | 2013-04-03 | 苏州大学 | Image retrieval method based on vocabulary tree level semantic model |
| CN106339486A (en) * | 2016-08-30 | 2017-01-18 | 西安电子科技大学 | Image retrieval method based on incremental learning of large vocabulary tree |
| CN108647307A (en) * | 2018-05-09 | 2018-10-12 | 京东方科技集团股份有限公司 | Image processing method, device, electronic equipment and storage medium |
| CN108710690A (en) * | 2018-05-22 | 2018-10-26 | 长春师范大学 | Medical image search method based on geometric verification |
-
2019
- 2019-06-27 CN CN201910567616.XA patent/CN110263198A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20090303511A1 (en) * | 2005-12-19 | 2009-12-10 | Xerox Corproration | Database system for processing image data for consistent color printing |
| CN103020111A (en) * | 2012-10-29 | 2013-04-03 | 苏州大学 | Image retrieval method based on vocabulary tree level semantic model |
| CN106339486A (en) * | 2016-08-30 | 2017-01-18 | 西安电子科技大学 | Image retrieval method based on incremental learning of large vocabulary tree |
| CN108647307A (en) * | 2018-05-09 | 2018-10-12 | 京东方科技集团股份有限公司 | Image processing method, device, electronic equipment and storage medium |
| CN108710690A (en) * | 2018-05-22 | 2018-10-26 | 长春师范大学 | Medical image search method based on geometric verification |
Non-Patent Citations (3)
| Title |
|---|
| WEN-CHENG LIN: "Integrating textual and visual information for cross-language image retrieval: A trans-media dictionary approach", 《INFORMATION PROCESSING & MANAGEMENT》 * |
| 张星宇: "基于K中心算法分层视觉词典树构造方法", 《四川理工学院学报(自然科学版)》 * |
| 李卫疆: "《基于自然语言处理的信息检索》", 31 July 2014, 云南大学出版社 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111191067A (en) * | 2019-12-25 | 2020-05-22 | 深圳市优必选科技股份有限公司 | Picture book identification method, terminal device and computer readable storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107291871B (en) | Matching degree evaluation method, device and medium for multi-domain information based on artificial intelligence | |
| Fang et al. | From captions to visual concepts and back | |
| JP5134628B2 (en) | Media material analysis of consecutive articles | |
| CN105224521B (en) | Key phrases extraction method and the method and device using its acquisition correlated digital resource | |
| US8503769B2 (en) | Matching text to images | |
| CN108829893A (en) | Determine method, apparatus, storage medium and the terminal device of video tab | |
| JP2003167914A (en) | Multimedia information retrieving method, program, recording medium and system therefor | |
| CN105849720A (en) | Visual semantic complex network and method for forming network | |
| WO2014050774A1 (en) | Document classification assisting apparatus, method and program | |
| WO2016057984A1 (en) | Methods and systems for base map and inference mapping | |
| CN104537028B (en) | A kind of Web information processing method and device | |
| CN109933660A (en) | The API information search method based on handout and Stack Overflow towards natural language form | |
| CN112836487A (en) | Automatic comment method and device, computer equipment and storage medium | |
| JP6420268B2 (en) | Image evaluation learning device, image evaluation device, image search device, image evaluation learning method, image evaluation method, image search method, and program | |
| CN103853797B (en) | A kind of picture retrieval method and system based on n member picture indices structures | |
| JP3820878B2 (en) | Information search device, score determination device, information search method, score determination method, and program recording medium | |
| JP2011248596A (en) | Searching system and searching method for picture-containing documents | |
| KR20180129001A (en) | Method and System for Entity summarization based on multilingual projected entity space | |
| CN105354182B (en) | The method and the method and device using its generation special topic for obtaining correlated digital resource | |
| Lokoč et al. | Revisiting SIRET video retrieval tool | |
| CN112148886A (en) | Method and system for constructing content knowledge graph | |
| CN106599072A (en) | Text clustering method and device | |
| CN110263198A (en) | A kind of search method and device | |
| KR101841615B1 (en) | Apparatus and method for computing noun similarities using semantic contexts | |
| JP2007317133A (en) | Image classification method, device, and program |
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 | ||
| CB02 | Change of applicant information |
Address after: 230088 China (Anhui) pilot Free Trade Zone, Hefei, Anhui province 6 / F and 23 / F, scientific research building, building 2, zone a, China sound Valley, No. 3333 Xiyou Road, high tech Zone, Hefei Applicant after: Anhui taoyun Technology Co.,Ltd. Address before: 230088 9th floor, building 1, tianyuandike science and Technology Park, 66 Qianshui East Road, high tech Zone, Hefei City, Anhui Province Applicant before: ANHUI TAOYUN TECHNOLOGY Co.,Ltd. |
|
| CB02 | Change of applicant information | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190920 |
|
| RJ01 | Rejection of invention patent application after publication |