CN107247730A - Image searching method and device - Google Patents
Image searching method and device Download PDFInfo
- Publication number
- CN107247730A CN107247730A CN201710309031.9A CN201710309031A CN107247730A CN 107247730 A CN107247730 A CN 107247730A CN 201710309031 A CN201710309031 A CN 201710309031A CN 107247730 A CN107247730 A CN 107247730A
- Authority
- CN
- China
- Prior art keywords
- picture
- search
- preset
- feature vector
- preset dimension
- 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 42
- 239000013598 vector Substances 0.000 claims abstract description 167
- 238000013135 deep learning Methods 0.000 claims abstract description 53
- 238000004364 calculation method Methods 0.000 claims abstract description 19
- 230000000007 visual effect Effects 0.000 claims description 19
- 230000002085 persistent effect Effects 0.000 claims description 18
- 238000010586 diagram Methods 0.000 description 3
- 229920000742 Cotton Polymers 0.000 description 1
- 241000283074 Equus asinus Species 0.000 description 1
- 241000282326 Felis catus Species 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 238000002372 labelling Methods 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 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
- G06F16/58—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
Landscapes
- Engineering & Computer Science (AREA)
- Library & Information Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Databases & Information Systems (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of image searching method and device, image searching method includes：Under the deep learning network model trained, the characteristic vector of the preset dimension of picture is obtained, the characteristic vector of the preset dimension is used for the perceptual property for reflecting picture comprehensively, and the picture includes the search pictures in Target Photo and picture database；Calculate the Euclidean distance of the characteristic vector of the characteristic vector of the preset dimension of search pictures and the preset dimension of Target Photo；Meet the search pictures of preparatory condition in output picture database, the preparatory condition is that the characteristic vector of the preset dimension of search pictures and the Euclidean distance of characteristic vector of preset dimension of Target Photo are less than first threshold.The present invention improves search efficiency while amount of calculation is taken into account with search precision.
Description
Technical Field
The present invention relates to the field of image processing technologies, and in particular, to a method and an apparatus for searching for a picture.
Background
In the field of video data, such as the deduplication of short videos, it is necessary to search hundreds of millions of pictures for similar pictures. At present, many prior arts describe methods for searching for similarity of pictures, but only comparing whether two pictures are similar requires a lot of calculation, and cannot be used in scenes with hundreds of millions of pictures. However, some algorithms with small calculation amount, such as extracting image fingerprints, lose too much image characteristic information in the process of searching for image similarity, and cause poor search results. Therefore, how to improve the search efficiency while simultaneously considering both the amount of calculation and the search accuracy is a problem that needs to be solved urgently.
Disclosure of Invention
The invention mainly aims to provide an image searching method, aiming at improving the searching efficiency while considering both the calculated amount and the searching precision.
In order to achieve the above object, the image searching method provided by the present invention comprises the following steps:
acquiring a feature vector of a preset dimension of a picture under a trained deep learning network model, wherein the feature vector of the preset dimension is used for comprehensively reflecting the visual attribute of the picture, and the picture comprises a target picture and a search picture in a picture database;
calculating Euclidean distance between the feature vector of the preset dimension of the search picture and the feature vector of the preset dimension of the target picture;
outputting the search picture which meets a preset condition in a picture database, wherein the preset condition is that the Euclidean distance between a feature vector of a preset dimension of the search picture and a feature vector of a preset dimension of the target picture is smaller than a first threshold value.
Preferably, before the step of obtaining the feature vector of the preset dimension of the picture under the trained deep learning network model, the method further includes:
calculating parameters of each layer in the deep learning network model by adopting an existing training atlas; wherein,
under the trained deep learning network model, acquiring the feature vector of the preset dimension of the picture comprises the following steps:
substituting parameters of each layer in the deep learning network model into the deep learning network model, and storing the parameters of each layer in the deep learning network model to obtain a persistent network model;
and extracting the characteristics of the target picture and the search picture by adopting the persistent network model to obtain the characteristic vectors of the preset dimensions of the target picture and the search picture.
Preferably, the step of calculating the euclidean distance between the feature vector of the preset dimension of the search picture and the feature vector of the preset dimension of the target picture includes:
storing the feature vectors of the preset dimensionality of the search pictures in the picture database and the corresponding file names of the search pictures in a preset area to form a search set;
establishing a data index for the data of the search set;
deleting the search picture corresponding to the feature vector of the preset dimension of which the Euclidean distance from the feature vector of the preset dimension of the target picture is greater than a second threshold value in the search set according to the data index to form a candidate set;
and calculating the Euclidean distance between the feature vector of the preset dimension of the search picture in the candidate set and the feature vector of the preset dimension of the target picture.
Preferably, the method for indexing data of the search set comprises VA-File or X-Tree.
Preferably, the preset-dimension feature vector is a 2048-dimension feature vector.
In addition, to achieve the above object, the present invention also provides an image search apparatus comprising:
the acquisition module is used for acquiring a feature vector of a preset dimension of a picture under a trained deep learning network model, wherein the feature vector of the preset dimension is used for comprehensively reflecting the visual attribute of the picture, and the picture comprises a target picture and a search picture in a picture database;
the first calculation module is used for calculating the Euclidean distance between the feature vector of the preset dimension of the search picture and the feature vector of the preset dimension of the target picture;
the output module is used for outputting the search picture which meets a preset condition in a picture database, wherein the preset condition is that the Euclidean distance between the feature vector of the preset dimension of the search picture and the feature vector of the preset dimension of the target picture is smaller than a first threshold value.
Preferably, the picture search apparatus further comprises a second calculation module,
the second calculation module is used for calculating parameters of each layer in the deep learning network model by adopting the existing training atlas;
the obtaining module is specifically used for substituting parameters of each layer in the deep learning network model into the deep learning network model, storing the parameters of each layer in the deep learning network model to obtain a persistent network model, and extracting features of the picture according to the persistent network model to obtain a feature vector of a preset dimension of the picture.
Preferably, the first calculation module comprises:
the storage unit is used for storing the feature vectors of the preset dimensionality of the search pictures in the picture database and the corresponding file names of the search pictures into a preset area to form a search set;
the data index establishing unit is used for establishing a data index divided by space for the data of the search set according to the data distribution condition of the search set;
the processing unit is used for deleting the search pictures corresponding to the feature vectors of the preset dimensionality with Euclidean distance from the feature vectors of the preset dimensionality of the target picture larger than a second threshold value in the search set according to the data index to form a candidate set;
a calculating unit, configured to calculate the euclidean distance between the feature vector of the preset dimension of the search picture in the candidate set and the feature vector of the preset dimension of the target picture.
Preferably, the method for indexing data of the search set comprises VA-File or X-Tree.
Preferably, the preset-dimension feature vector is a 2048-dimension feature vector.
The method comprises the steps of obtaining a feature vector of a preset dimension of a picture under a trained deep learning network model; calculating the Euclidean distance between the feature vector of the preset dimension of the search picture and the feature vector of the preset dimension of the target picture; and outputting the search picture which meets preset conditions in the picture database, wherein the preset conditions are that the Euclidean distance between the characteristic vector of the preset dimensionality of the search picture and the characteristic vector of the preset dimensionality of the target picture is smaller than a first threshold value. According to the trained deep learning network model, the characteristic vector of the preset dimension representing the visual attribute of the picture is obtained, and then the Euclidean distance between the search picture and the characteristic vector of the preset dimension of the target picture is judged to detect whether the pictures are similar or not, so that the search efficiency is improved while the calculation amount and the search precision are considered.
Drawings
FIG. 1 is a flowchart illustrating a first embodiment of an image searching method according to the present invention;
FIG. 2 is a flowchart illustrating a second embodiment of an image searching method according to the present invention;
FIG. 3 is a flowchart illustrating a third embodiment of an image searching method according to the present invention;
FIG. 4 is a schematic structural diagram of a first embodiment of an image search apparatus according to the present invention;
FIG. 5 is a diagram illustrating a second embodiment of an image searching apparatus according to the present invention;
fig. 6 is a schematic diagram of a detailed structure of the first calculating module in the third embodiment of the image searching apparatus according to the present invention.
The implementation, functional features and advantages of the objects of the present invention will be further explained with reference to the accompanying drawings.
Detailed Description
It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
The invention provides an image searching method, and referring to fig. 1, in an embodiment, the image searching method includes:
step S10, acquiring a feature vector of a preset dimension of a picture under a trained deep learning network model, wherein the feature vector of the preset dimension is used for comprehensively reflecting the visual attribute of the picture, and the picture comprises a target picture and a search picture in a picture database;
the picture searching method provided by the invention is mainly applied to an image processing system and used for searching similar pictures in a picture database, and the method can simultaneously take account of calculated amount and searching precision and can search out similar pictures quickly.
Specifically, searching for similar pictures is realized, and firstly, visual attributes capable of comprehensively reflecting the pictures need to be extracted, in this embodiment, a feature vector of a preset dimension is extracted to represent the view angle attributes of the pictures, and the feature vector of the preset dimension is obtained through a trained deep learning network model.
Step S20, calculating the euclidean distance between the feature vector of the preset dimension of the search picture and the feature vector of the preset dimension of the target picture;
step S30, outputting the search picture that meets a preset condition in the picture database, where the preset condition is that the euclidean distance between the feature vector of the preset dimension of the search picture and the feature vector of the preset dimension of the target picture is smaller than a first threshold.
The feature vectors of the preset dimensions of the picture can be set according to actual needs, the definition of the picture is insufficient due to too small dimensions, and the calculated amount is too large due to too large dimensions, and in order to take account of the calculated amount and the search precision, in this embodiment, 2048 dimensions will be taken as an example for detailed description. The first threshold may be set according to actual needs, and if the first threshold is too large, the search accuracy may be insufficient, many dissimilar pictures may be mistaken for similar pictures to be output, and if the first threshold is too small, some similar pictures may not be searched. It is understood that "5" in "the first threshold is 5" means 5 unit vector lengths. In processing vector calculations, the distance between vectors is measured by the modulus of the unit vector, including the euclidean distance mentioned in this application. A unit vector is defined as a vector modulo equal to 1, i.e. a vector of length 1. That is, the distance between the vectors is characterized by 1, and the euclidean distance mentioned in this application is also characterized by 1. For example, the euclidean distance between two vectors is calculated to be 3.4 unit vector lengths, which can be directly expressed as: the euclidean distance of the two vectors is 3.4.
Specifically, under a trained deep learning network model, 2048-dimensional feature vectors of a target picture and search pictures in a picture database are obtained, the feature vectors are enough to represent visual attributes of the pictures, Euclidean distances between the 2048-dimensional feature vectors of all the search pictures in the picture database and the 2048-dimensional feature vectors of the target picture are calculated, when the Euclidean distances are smaller than 5, the search pictures are considered to be similar to the target picture, and all the search pictures of which the Euclidean distances are smaller than 5 in the picture database are output.
It can be understood that the searching for the picture similar to the target picture by searching for the euclidean distance between the feature vector of the preset dimension of the search picture in the picture database and the feature vector of the preset dimension of the target picture is less than the first threshold value is because: the characteristic vector of the preset dimension of the picture can comprehensively reflect the visual attribute of the picture. Visual attributes are defined as the inherent visual characteristics of an object, including visual attributes with semantic meaning and visual attributes with non-semantic meaning. Semantic attributes are textual descriptions that can be learned through manual labeling for describing and identifying objects. As described for the object components, with nose, with legs, with feet, etc.; describing the shape of the object, circular, square, triangular, etc.; describing the components of the object, it is of plush, silk, cotton, etc. that describes the material of the object. Non-semantic attributes refer to textual descriptions that are used to distinguish object classes, such as cat does not, dog does, horse does not, donkey does, etc. If the feature vectors of the preset dimensionality of the two pictures are completely the same, namely the Euclidean distance is 0, the two pictures can be considered to be completely the same, and therefore, when the Euclidean distance between the feature vector of the preset dimensionality of the search picture in the picture database and the feature vector of the preset dimensionality of the target picture is smaller than a certain value, the search picture can be considered to be similar to the target picture.
The embodiment of the invention obtains the characteristic vector of the preset dimensionality of the picture under the trained deep learning network model; calculating the Euclidean distance between the feature vector of the preset dimension of the search picture and the feature vector of the preset dimension of the target picture; and outputting the search picture which meets preset conditions in the picture database, wherein the preset conditions are that the Euclidean distance between the characteristic vector of the preset dimensionality of the search picture and the characteristic vector of the preset dimensionality of the target picture is smaller than a first threshold value. According to the trained deep learning network model, the characteristic vector of the preset dimension representing the visual attribute of the picture is obtained, and then the Euclidean distance between the search picture and the characteristic vector of the preset dimension of the target picture is judged to detect whether the pictures are similar or not, so that the search efficiency is improved while the calculation amount and the search precision are considered.
Further, referring to fig. 2, a second embodiment of the picture searching method of the present invention is provided, and based on the first embodiment of the picture searching method of the present invention, before the step S10, the second embodiment of the picture searching method further includes:
step S40, calculating parameters of each layer in the deep learning network model by adopting the existing training atlas; wherein,
the step S10 includes:
step S11, substituting the parameters of each layer in the deep learning network model into the deep learning network model, and storing the parameters of each layer in the deep learning network model to obtain a persistent network model;
and step S12, extracting the features of the target picture and the search picture by adopting the persistent network model to obtain the feature vectors of the preset dimensions of the target picture and the search picture.
Specifically, an existing training atlas is adopted to train the deep learning network model, parameters of each layer in the deep learning network model are calculated, and the parameters are stored in a local disk. And substituting the parameters of each layer in the deep learning network model into the deep learning network model to obtain the persistent network model. And then removing a classification layer in the persistent network model, taking the second last layer used for extracting the picture feature vector in the persistent network model as output, and extracting the feature vectors with preset dimensions from the target picture and all the search pictures in the picture database to obtain the feature vectors with the preset dimensions of the target picture and the search pictures.
Further, referring to fig. 3, a third embodiment of the picture searching method of the present invention is provided, and based on the first embodiment of the picture searching method of the present invention, in the third embodiment of the picture searching method, step S20 includes:
step S21, storing the feature vectors of the preset dimensionality of the search pictures in the picture database and the corresponding file names of the search pictures into a preset area to form a search set;
step S22, establishing data index for the data of the search set;
step S23, deleting the search set according to the data index, and forming a candidate set by the search picture corresponding to the feature vector of the preset dimension of which the Euclidean distance from the feature vector of the preset dimension of the target picture is greater than a second threshold value;
step S24, calculating the euclidean distance between the feature vector of the preset dimension of the search picture in the candidate set and the feature vector of the preset dimension of the target picture.
In this embodiment, when there are too many search pictures in the picture database, in order to reduce the calculation amount of the euclidean distance, the picture with the excessively large euclidean distance with respect to the feature vector of the preset dimension of the target picture is first deleted, and then the euclidean distance between the feature vector of the preset dimension of the remaining search pictures in the picture database (i.e., the feature vector of the preset dimension of the candidate set search picture) and the feature vector of the preset dimension of the target picture is calculated. Specifically, the data of the search set may be indexed by using a VA-File (Vector Approximation File) or X-Tree (X-Tree) method, that is, the method for establishing a data index divided in space for the data of the search set according to the data distribution condition of the search set includes the VA-File or X-Tree, and it is understood that in other embodiments, other methods may also be used to establish an index for the data of the search set.
Referring to fig. 4, in an embodiment, the image searching apparatus of the present invention includes:
the acquisition module 10 is configured to acquire a feature vector of a preset dimension of a picture under a trained deep learning network model, where the feature vector of the preset dimension is used to comprehensively reflect visual attributes of the picture, and the picture includes a target picture and a search picture in a picture database;
the picture searching device provided by the invention is mainly applied to an image processing system and used for searching similar pictures in a picture database.
Specifically, searching for similar pictures is realized, and firstly, visual attributes capable of comprehensively reflecting the pictures need to be extracted, in this embodiment, a feature vector of a preset dimension is extracted to represent the view angle attributes of the pictures, and the feature vector of the preset dimension is obtained through a trained deep learning network model.
A first calculating module 20, configured to calculate a euclidean distance between the feature vector of the preset dimension of the search picture and the feature vector of the preset dimension of the target picture;
the output module 30 is configured to output the search picture that meets a preset condition in a picture database, where the preset condition is that the euclidean distance between the feature vector of the preset dimension of the search picture and the feature vector of the preset dimension of the target picture is smaller than a first threshold.
The feature vectors of the preset dimensions of the picture can be set according to actual needs, the definition of the picture is insufficient due to too small dimensions, and the calculated amount is too large due to too large dimensions, and in order to take account of the calculated amount and the search precision, in this embodiment, 2048 dimensions will be taken as an example for detailed description. The first threshold may be set according to actual needs, and if the first threshold is too large, the search accuracy may be insufficient, many dissimilar pictures may be mistaken for similar pictures to be output, and if the first threshold is too small, some similar pictures may not be searched.
Specifically, under a trained deep learning network model, 2048-dimensional feature vectors of a target picture and search pictures in a picture database are obtained, the feature vectors are enough to represent visual attributes of the pictures, Euclidean distances between the 2048-dimensional feature vectors of all the search pictures in the picture database and the 2048-dimensional feature vectors of the target picture are calculated, when the Euclidean distances are smaller than 5, the search pictures are considered to be similar to the target picture, and all the search pictures of which the Euclidean distances are smaller than 5 in the picture database are output.
It can be understood that the searching for the picture similar to the target picture by searching for the euclidean distance between the feature vector of the preset dimension of the search picture in the picture database and the feature vector of the preset dimension of the target picture is less than the first threshold value is because: the characteristic vectors of the preset dimensionalities of the pictures can comprehensively reflect the visual attributes of the pictures, if the characteristic vectors of the preset dimensionalities of the two pictures are completely the same, namely the Euclidean distance is 0, the two pictures can be considered to be completely the same, and therefore when the Euclidean distance between the characteristic vector of the preset dimensionality of the search picture in the picture database and the characteristic vector of the preset dimensionality of the target picture is smaller than a certain value, the search picture can be considered to be similar to the target picture.
The embodiment of the invention obtains the characteristic vector of the preset dimensionality of the picture under the trained deep learning network model; calculating the Euclidean distance between the feature vector of the preset dimension of the search picture and the feature vector of the preset dimension of the target picture; and outputting the search picture which meets preset conditions in the picture database, wherein the preset conditions are that the Euclidean distance between the characteristic vector of the preset dimensionality of the search picture and the characteristic vector of the preset dimensionality of the target picture is smaller than a first threshold value. According to the trained deep learning network model, the characteristic vector of the preset dimension representing the visual attribute of the picture is obtained, and then the Euclidean distance between the search picture and the characteristic vector of the preset dimension of the target picture is judged to detect whether the pictures are similar or not, so that the search efficiency is improved while the calculation amount and the search precision are considered.
Further, referring to fig. 5, a second embodiment of the picture search apparatus of the present invention is provided, in which, based on the first embodiment of the picture search apparatus of the present invention, the picture search apparatus further comprises a second calculation module 40,
the second calculating module 40 is configured to calculate parameters of each layer in the deep learning network model by using an existing training atlas;
the obtaining module 10 is specifically configured to substitute parameters of each layer in the deep learning network model into the deep learning network model, store the parameters of each layer in the deep learning network model to obtain a persistent network model, and perform feature extraction on the picture by using the persistent network model to obtain feature vectors of preset dimensions of the target picture and the search picture.
Specifically, an existing training atlas is adopted to train the deep learning network model, parameters of each layer in the deep learning network model are calculated, and the parameters are stored in a local disk. And substituting the parameters of each layer in the deep learning network model into the deep learning network model to obtain the persistent network model. And then removing a classification layer in the persistent network model, taking the second last layer used for extracting the picture feature vector in the persistent network model as output, and extracting the feature vectors with preset dimensions from the target picture and all the search pictures in the picture database to obtain the feature vectors with the preset dimensions of the target picture and the search pictures.
Further, referring to fig. 6, a third embodiment of the image searching apparatus of the present invention is provided, based on the first embodiment of the image searching apparatus of the present invention, in the third embodiment of the image searching apparatus, the first calculating module 20 includes:
the storage unit 21 is configured to store the feature vector of the preset dimension of the search picture obtained in the picture database and the corresponding file name of the search picture to a preset area, so as to form a search set;
a data index establishing unit 22, configured to establish a data index for the data of the search set;
the processing unit 23 is configured to delete the search picture corresponding to the feature vector of the preset dimension, where an euclidean distance between the search picture and the feature vector of the preset dimension of the target picture is greater than a second threshold, in the search set according to the data index, and form a candidate set;
a calculating unit 24, configured to calculate the euclidean distance between the feature vector of the preset dimension of the search picture in the candidate set and the feature vector of the preset dimension of the target picture.
In this embodiment, when there are too many search pictures in the picture database, in order to reduce the calculation amount of the euclidean distance, the picture with the excessively large euclidean distance with respect to the feature vector of the preset dimension of the target picture is first deleted, and then the euclidean distance between the feature vector of the preset dimension of the remaining search pictures in the picture database (i.e., the feature vector of the preset dimension of the candidate set search picture) and the feature vector of the preset dimension of the target picture is calculated. Specifically, the VA-File or X-Tree method may be used to index the data in the search set, that is, the method for establishing a spatially partitioned data index for the data in the search set according to the data distribution of the search set includes VA-File or X-Tree, and it is understood that in other embodiments, other methods may also be used to index the data in the search set.
The above description is only a preferred embodiment of the present invention, and not intended to limit the scope of the present invention, and all modifications of equivalent structures and equivalent processes, which are made by using the contents of the present specification and the accompanying drawings, or directly or indirectly applied to other related technical fields, are included in the scope of the present invention.
Claims (10)
1. An image searching method, characterized in that the image searching method comprises the following steps:
acquiring a feature vector of a preset dimension of a picture under a trained deep learning network model, wherein the feature vector of the preset dimension is used for comprehensively reflecting the visual attribute of the picture, and the picture comprises a target picture and a search picture in a picture database;
calculating Euclidean distance between the feature vector of the preset dimension of the search picture and the feature vector of the preset dimension of the target picture;
outputting the search picture which meets a preset condition in the picture database, wherein the preset condition is that the Euclidean distance between the characteristic vector of the preset dimension of the search picture and the characteristic vector of the preset dimension of the target picture is smaller than a first threshold value.
2. The method for searching pictures according to claim 1, wherein before the step of obtaining the feature vector of the preset dimension of the picture under the trained deep learning network model, the method further comprises:
calculating parameters of each layer in the deep learning network model by adopting an existing training atlas; wherein,
under the trained deep learning network model, acquiring the feature vector of the preset dimension of the picture comprises the following steps:
substituting the calculated parameters of each layer in the deep learning network model into the deep learning network model, and storing the parameters of each layer in the deep learning network model to obtain a persistent network model;
and extracting the characteristics of the target picture and the search picture by adopting the persistent network model to obtain the characteristic vectors of the preset dimensions of the target picture and the search picture.
3. The picture searching method according to claim 1, wherein the step of calculating the euclidean distance between the feature vector of the preset dimension of the search picture and the feature vector of the preset dimension of the target picture comprises:
storing the feature vectors of the preset dimensionality of the search pictures in the picture database and the corresponding file names of the search pictures into a preset area to form a search set;
establishing a data index for the data of the search set;
deleting the search picture corresponding to the feature vector of the preset dimension of which the Euclidean distance from the feature vector of the preset dimension of the target picture is greater than a second threshold value in the search set according to the data index to form a candidate set;
and calculating the Euclidean distance between the characteristic vector of the preset dimension of the search picture in the candidate set and the characteristic vector of the preset dimension of the target picture.
4. The picture searching method according to claim 3, wherein the method of indexing data of the search set comprises VA-File or X-Tree.
5. The picture searching method according to any one of claims 1 to 3, wherein the preset-dimension feature vector is a 2048-dimension feature vector.
6. An image search device, comprising:
the acquisition module is used for acquiring a feature vector of a preset dimension of a picture under a trained deep learning network model, wherein the feature vector of the preset dimension is used for comprehensively reflecting the visual attribute of the picture, and the picture comprises a target picture and a search picture in a picture database;
the first calculation module is used for calculating the Euclidean distance between the feature vector of the preset dimension of the search picture and the feature vector of the preset dimension of the target picture;
the output module is used for outputting the search picture which meets a preset condition in a picture database, wherein the preset condition is that the Euclidean distance between the feature vector of the preset dimension of the search picture and the feature vector of the preset dimension of the target picture is smaller than a first threshold value.
7. The picture search apparatus of claim 6, further comprising a second calculation module,
the second calculation module is used for calculating parameters of each layer in the deep learning network model by adopting the existing training atlas;
the acquisition module is specifically used for substituting parameters of each layer in the deep learning network model into the deep learning network model, storing the parameters of each layer in the deep learning network model to obtain a persistent network model, and extracting the features of the picture by adopting the persistent network model to obtain the feature vectors of the preset dimensions of the target picture and the search picture.
8. The picture search apparatus of claim 6, wherein the first calculation module comprises:
the storage unit is used for storing the feature vectors of the preset dimensionality of the search pictures in the picture database and the corresponding file names of the search pictures into a preset area to form a search set;
the data index establishing unit is used for establishing a data index for the data of the search set;
the processing unit is used for deleting the search set according to the data index, and the search picture corresponding to the feature vector of the preset dimension of the target picture, the Euclidean distance of which from the feature vector of the preset dimension of the target picture is greater than a second threshold value, so as to form a candidate set;
a calculating unit, configured to calculate the euclidean distance between the feature vector of the preset dimension of the search picture in the candidate set and the feature vector of the preset dimension of the target picture.
9. The picture searching apparatus according to claim 8, wherein the method of indexing data of the search set comprises VA-File or X-Tree.
10. The picture search apparatus according to any one of claims 6 to 8, wherein the predetermined-dimension feature vector is a 2048-dimension feature vector.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710309031.9A CN107247730A (en) | 2017-05-04 | 2017-05-04 | Image searching method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710309031.9A CN107247730A (en) | 2017-05-04 | 2017-05-04 | Image searching method and device |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107247730A true CN107247730A (en) | 2017-10-13 |
Family
ID=60016916
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710309031.9A Pending CN107247730A (en) | 2017-05-04 | 2017-05-04 | Image searching method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107247730A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107766492A (en) * | 2017-10-18 | 2018-03-06 | 北京京东尚科信息技术有限公司 | A kind of method and apparatus of picture search |
| CN108733780A (en) * | 2018-05-07 | 2018-11-02 | 浙江大华技术股份有限公司 | A kind of image searching method and device |
| CN110019870A (en) * | 2017-12-29 | 2019-07-16 | 浙江宇视科技有限公司 | The image search method and system of image cluster based on memory |
| CN110442738A (en) * | 2019-07-31 | 2019-11-12 | 北京明略软件系统有限公司 | Picture De-weight method and device, storage medium and electronic device |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104679863A (en) * | 2015-02-28 | 2015-06-03 | 武汉烽火众智数字技术有限责任公司 | Method and system for searching images by images based on deep learning |
| CN105808732A (en) * | 2016-03-10 | 2016-07-27 | 北京大学 | Integration target attribute identification and precise retrieval method based on depth measurement learning |
| CN106354735A (en) * | 2015-07-22 | 2017-01-25 | 杭州海康威视数字技术股份有限公司 | Image target searching method and device |
-
2017
- 2017-05-04 CN CN201710309031.9A patent/CN107247730A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104679863A (en) * | 2015-02-28 | 2015-06-03 | 武汉烽火众智数字技术有限责任公司 | Method and system for searching images by images based on deep learning |
| CN106354735A (en) * | 2015-07-22 | 2017-01-25 | 杭州海康威视数字技术股份有限公司 | Image target searching method and device |
| CN105808732A (en) * | 2016-03-10 | 2016-07-27 | 北京大学 | Integration target attribute identification and precise retrieval method based on depth measurement learning |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107766492A (en) * | 2017-10-18 | 2018-03-06 | 北京京东尚科信息技术有限公司 | A kind of method and apparatus of picture search |
| CN107766492B (en) * | 2017-10-18 | 2020-07-31 | 北京京东尚科信息技术有限公司 | Image searching method and device |
| CN110019870A (en) * | 2017-12-29 | 2019-07-16 | 浙江宇视科技有限公司 | The image search method and system of image cluster based on memory |
| CN108733780A (en) * | 2018-05-07 | 2018-11-02 | 浙江大华技术股份有限公司 | A kind of image searching method and device |
| WO2019214565A1 (en) * | 2018-05-07 | 2019-11-14 | Zhejiang Dahua Technology Co., Ltd. | Systems and methods for image searching |
| CN108733780B (en) * | 2018-05-07 | 2020-06-23 | 浙江大华技术股份有限公司 | Picture searching method and device |
| US11409984B2 (en) | 2018-05-07 | 2022-08-09 | Zhejiang Dahua Technology Co., Ltd. | Systems and methods for image searching |
| CN110442738A (en) * | 2019-07-31 | 2019-11-12 | 北京明略软件系统有限公司 | Picture De-weight method and device, storage medium and electronic device |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109815364B (en) | Method and system for extracting, storing and retrieving mass video features | |
| US8520909B2 (en) | Automatic and semi-automatic image classification, annotation and tagging through the use of image acquisition parameters and metadata | |
| CN106446816B (en) | Face recognition method and device | |
| CN109960742B (en) | Local information searching method and device | |
| US9076195B2 (en) | Methods and apparatus to identify components from images of the components | |
| CN105243060B (en) | A kind of method and device of retrieving image | |
| US20100250539A1 (en) | Shape based picture search | |
| CN110019891B (en) | Image storage method, image retrieval method and device | |
| US9633272B2 (en) | Real time object scanning using a mobile phone and cloud-based visual search engine | |
| CN107247730A (en) | Image searching method and device | |
| CN114168768A (en) | Image retrieval method and related equipment | |
| CN114491083B (en) | Knowledge graph library construction method and knowledge graph library construction device in industrial detection | |
| KR101472451B1 (en) | System and Method for Managing Digital Contents | |
| KR20150137006A (en) | Annotation display assistance device and method of assisting annotation display | |
| CN111222409A (en) | Vehicle brand labeling method, device and system | |
| US8655016B2 (en) | Example-based object retrieval for video surveillance | |
| JP2008117271A (en) | Object recognition device of digital image, program and recording medium | |
| US9762934B2 (en) | Apparatus and method for verifying broadcast content object identification based on web data | |
| CN110674342B (en) | Method and device for inquiring target image | |
| WO2017107361A1 (en) | Method and device for determining landscape information of picture | |
| CN114491130B (en) | Image retrieval method, device and computer-readable storage medium | |
| KR101111046B1 (en) | A Similar Video Search System through Object Detection Information and A Method thereof | |
| KR20080046490A (en) | Face Recognition Method and Device Using Montage Face Image | |
| CN118227850A (en) | Test question duplicate checking method and system | |
| CN110275990B (en) | Method and device for generating KV stored key and value |
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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20171013 |