CN106844715A - A kind of picture retrieval matching process and device - Google Patents
A kind of picture retrieval matching process and device Download PDFInfo
- Publication number
- CN106844715A CN106844715A CN201710068557.2A CN201710068557A CN106844715A CN 106844715 A CN106844715 A CN 106844715A CN 201710068557 A CN201710068557 A CN 201710068557A CN 106844715 A CN106844715 A CN 106844715A
- Authority
- CN
- China
- Prior art keywords
- picture
- eigenmatrix
- module
- characteristic vector
- feature
- 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
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
- G06F16/5838—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content using colour
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/74—Image or video pattern matching; Proximity measures in feature spaces
- G06V10/75—Organisation of the matching processes, e.g. simultaneous or sequential comparisons of image or video features; Coarse-fine approaches, e.g. multi-scale approaches; using context analysis; Selection of dictionaries
- G06V10/757—Matching configurations of points or features
Abstract
The invention discloses a kind of picture retrieval matching process and device, method includes:S101, the picture to be matched for receiving user input;S102, the local feature region for extracting picture to be matched;S103, the eigenmatrix that picture to be matched is calculated according to the local feature region of picture to be matched;S104, the eigenmatrix of picture to be matched is carried out into vectorization as input feature vector matrix it is converted to characteristic vector;S105, the characteristic vector of picture to be matched is matched with the characteristic vector in characteristic vector data storehouse, and pictorial information corresponding to the characteristic vector that will match to returns to user.Matching efficiency can be greatly improved in retrieval by the present invention, simultaneously because data volume is less, more data can be stored so that the result of retrieval matching is more accurate.
Description
Technical field
The present invention relates to picture searching field, and in particular to a kind of picture retrieval matching process and device.
Background technology
With the emergence of increasing electric business enterprise and platform, net purchase has turned into a kind of indispensable online purchase of people
Object space formula.But when being related to information search on some electric business platforms, most people are still to be entered using the mode of words input instantly
Row information is searched for, and as increasing electric business platform capital construction in the future is on mobile device, and user is more and more acceptant
Way of search based on picture carries out information search.
Instantly some large-scale electric business platforms progressively start to focus on based on picture during the function that constantly becomes more meticulous,
Information search.And due to the complexity of image content, instantly existing plurality of picture Information-retrieval Algorithm it is efficient with it is accurate
It is still a bottleneck for influence pictorial information retrieval.
The content of the invention
In order to overcome the deficiencies in the prior art, it is an object of the invention to provide a kind of picture retrieval matching process and dress
Put, realize the degree of accuracy of the efficiency and matching result of raising picture retrieval.
To solve the above problems, the technical solution adopted in the present invention is as follows:
Scheme one:
A kind of picture retrieval matching process, comprises the following steps:
S101, the picture to be matched for receiving user input;
S102, the local feature region for extracting picture to be matched;
S103, the eigenmatrix that picture to be matched is calculated according to the local feature region of picture to be matched;
S104, the eigenmatrix of picture to be matched is carried out into vectorization as input feature vector matrix it is converted to characteristic vector;
S105, the characteristic vector of picture to be matched is matched with the characteristic vector in characteristic vector data storehouse, and will matching
To characteristic vector corresponding to pictorial information return to user.
Preferably, the vectorization conversion is comprised the following steps:
A, the eigenmatrix for reading all pictures in picture database are simultaneously merged and obtain eigenmatrix M;
B, the random central feature value for generating predetermined number in the characteristic value space of eigenmatrix M;
C, the every a line for taking out eigenmatrix M, are designated as Fi, calculate successively and FiThe central feature value of arest neighbors;
D, each central feature value is updated successively, the rule of renewal is:It is special that current central feature value is updated to current center
The F of value indicative and arest neighborsiGeometric center value;
E, when judging that each central feature value updates produced displacement whether less than predetermined threshold value, if so, then perform step F,
If it is not, then return to step C;
Each central feature value after F, preservation renewal, forms a central feature vector;
G, input feature vector matrix is designated as M1, the every a line for M1 therefrom takes out corresponding arest neighbors in heart characteristic vector successively
Central feature value;
H, the hit-count for counting each central feature value being removed and obtain correspond to the input feature vector matrix feature
Vector.
Preferably, the characteristic vector in characteristic vector data storehouse is generated by following steps:
Step one, for putting all pictures in picture database into queue in;
Step 2, the local feature region for extracting every pictures in queue successively;
Step 3, the eigenmatrix for calculating every pictures according to the local feature region per pictures;
Step 4, it is converted to feature for the eigenmatrix of every pictures to be carried out into vectorization as input feature vector matrix successively
Vector, and corresponding characteristic vector will be preserved into characteristic vector data storehouse per pictures.
Scheme two:
A kind of picture retrieval coalignment, including with lower module:
Receiver module:Picture to be matched for receiving user input;
Feature point extraction module:Local feature region for extracting picture to be matched;
Eigenmatrix computing module:Feature square for calculating picture to be matched according to the local feature region of picture to be matched
Battle array;
Vectorization module:Spy is converted to for the eigenmatrix of picture to be matched to be carried out into vectorization as input feature vector matrix
Levy vector;
Matching returns to module:For the characteristic vector in the characteristic vector of picture to be matched and characteristic vector data storehouse to be carried out
Match somebody with somebody, and pictorial information corresponding to the characteristic vector that will match to returns to user.
Preferably, the vectorization conversion is included with lower module:
Merging module:Eigenmatrix M is obtained for reading the eigenmatrix of all pictures in picture database and merging;
Central feature value generation module:Center for the random generation predetermined number in the characteristic value space of eigenmatrix M is special
Value indicative;
First computing module:Every a line for taking out eigenmatrix M, is designated as Fi, calculate successively and FiThe center of arest neighbors
Characteristic value;
Central feature value update module:For updating each central feature value successively, the rule of renewal is:Current central feature
Value is updated to the F of current central feature value and arest neighborsiGeometric center value;
Judge module:For whether judging displacement produced during the renewal of each central feature value less than predetermined threshold value, if so, then
Implementation center's feature vector generation module, if it is not, then returning to the first computing module;
Central feature vector generation module:For preserving each the central feature value after updating, a central feature vector is formed;
Central feature value determining module:For input feature vector matrix to be designated as into M1, for every a line therefrom heart characteristic vector of M1
In take out the central feature value of corresponding arest neighbors successively;
Characteristic vector output module:For counting the hit-count of each central feature value being removed and obtaining corresponding to described
The characteristic vector of input feature vector matrix.
Preferably, the characteristic vector in characteristic vector data storehouse is generated by with lower module:
Module one, for putting all pictures in picture database into queue in;
Module two, the local feature region for extracting every pictures in queue successively;
Module three, the eigenmatrix for calculating every pictures according to the local feature region per pictures;
Module four, it is converted to feature for the eigenmatrix of every pictures to be carried out into vectorization as input feature vector matrix successively
Vector, and corresponding characteristic vector will be preserved into characteristic vector data storehouse per pictures.
Compared to existing technology, the beneficial effects of the present invention are:
Matching efficiency can be greatly improved in retrieval, simultaneously because data volume is less, more data can be stored so that
The result for retrieving matching is more accurate.
Brief description of the drawings
Fig. 1 is the flow chart of picture retrieval matching process of the invention.
Specific embodiment
Below, with reference to accompanying drawing and specific embodiment, the present invention is described further:
With reference to Fig. 1, a kind of picture retrieval matching process is comprised the following steps:
S101, the picture to be matched for receiving user input.
S102, the local feature region for extracting picture to be matched.
S103, the eigenmatrix that picture to be matched is calculated according to the local feature region of picture to be matched.
S104, using the eigenmatrix of picture to be matched as input feature vector matrix carry out vectorization be converted to feature to
Amount.
S105, the characteristic vector of picture to be matched is matched with the characteristic vector in characteristic vector data storehouse, and will
The pictorial information corresponding to characteristic vector for matching returns to user.Pictorial information is to be marked for per pictures in advance
Information.
Wherein, step S102 and S103 can realize that existing picture retrieval is also based on feature square using prior art
Battle array is retrieved, and because the data volume of eigenmatrix is big, needs to expend a large amount of internal memories during retrieval, therefore cannot be applied to extensive
Picture retrieval.Therefore, the present invention carries out vectorization conversion by by eigenmatrix, is converted to the characteristic vector of low latitudes, and
Data in characteristic vector are far smaller than the data in eigenmatrix, are so greatly reduced per the corresponding characteristic of pictures,
Matching efficiency during retrieval is also higher, and, server just can store more pictures, so as to support that large-scale picture is examined
Rope, the result of retrieval is more accurate.
Vectorization conversion principle be similar to clustering algorithm, the characteristic value to picture carries out polymerization treatment, specifically include with
Lower step:
A, the eigenmatrix for reading all pictures in picture database are simultaneously merged and obtain eigenmatrix M.
B, the random central feature value for generating predetermined number in the characteristic value space of eigenmatrix M.Wherein, predetermined number
Picture number in picture database determines that general picture has 100,000, then predetermined number is 10000, i.e., random generation
10000 central feature values.
C, the every a line for taking out eigenmatrix M, are designated as Fi, calculate successively and FiThe central feature value of arest neighbors.Feature square
Battle array M is equivalent to three-dimensional matrice, its eigenmatrix that a pictures are represented per a line.Calculate successively and FiThe center of arest neighbors is special
Value indicative refers to since the first row, to calculate the central feature value from the first row arest neighbors, from the central feature of the second row arest neighbors
Value, by that analogy.
D, each central feature value is updated successively, the rule of renewal is:Current central feature value be updated to it is current in
The F of heart characteristic value and arest neighborsiGeometric center value.Calculate two geometric center values of characteristic value special as new center
Value indicative.
E, when judging that each central feature value updates produced displacement whether less than predetermined threshold value, if so, then performing step
Rapid F, if it is not, then return to step C.
Each central feature value after F, preservation renewal, forms a central feature vector.
G, input feature vector matrix is designated as M1, for M1 every a line therefrom take out successively in heart characteristic vector it is corresponding most
The central feature value of neighbour.I.e. since the first row of M1, the central feature value with current line arest neighbors is sequentially found.
H, the hit-count for counting each central feature value being removed simultaneously obtain corresponding to the input feature vector matrix
Characteristic vector.
For example, central feature vector is designated as into CV=[C1, C2,…,C10000,], in steph, if hit C13, C25
It is secondary ..., Cn7 times, then the characteristic vector for obtaining is Cout=[3,5,…,7].Wherein, how many central feature value is removed, then
The characteristic vector for obtaining is with regard to how many dimension.
Generated by following steps for the characteristic vector in characteristic vector data storehouse:
Step one, for putting all pictures in picture database into queue in;
Step 2, the local feature region for extracting every pictures in queue successively;
Step 3, the eigenmatrix for calculating every pictures according to the local feature region per pictures;
Step 4, it is converted to feature for the eigenmatrix of every pictures to be carried out into vectorization as input feature vector matrix successively
Vector, and corresponding characteristic vector will be preserved into characteristic vector data storehouse per pictures.
By above-mentioned picture retrieval matching process, matching efficiency can be greatly improved in retrieval, simultaneously because data
That measures is less, can store more data so that the result of retrieval matching is more accurate.
Operational effect is as follows on 8G internal memories, 4 core servers:
Conventional method | The present invention | |
Storage picture number | 1950 | 22576 |
Single search time | 0.25s | 0.10s |
It is the invention also discloses a kind of picture retrieval coalignment including following corresponding to above-mentioned picture retrieval matching process
Module:
Receiver module:Picture to be matched for receiving user input;
Feature point extraction module:Local feature region for extracting picture to be matched;
Eigenmatrix computing module:Feature square for calculating picture to be matched according to the local feature region of picture to be matched
Battle array;
Vectorization module:Spy is converted to for the eigenmatrix of picture to be matched to be carried out into vectorization as input feature vector matrix
Levy vector;
Matching returns to module:For the characteristic vector in the characteristic vector of picture to be matched and characteristic vector data storehouse to be carried out
Match somebody with somebody, and pictorial information corresponding to the characteristic vector that will match to returns to user.
Preferably, the vectorization conversion is included with lower module:
Merging module:Eigenmatrix M is obtained for reading the eigenmatrix of all pictures in picture database and merging;
Central feature value generation module:Center for the random generation predetermined number in the characteristic value space of eigenmatrix M is special
Value indicative;
First computing module:Every a line for taking out eigenmatrix M, is designated as Fi, calculate successively and FiThe center of arest neighbors
Characteristic value;
Central feature value update module:For updating each central feature value successively, the rule of renewal is:Current central feature
Value is updated to the F of current central feature value and arest neighborsiGeometric center value;
Judge module:For whether judging displacement produced during the renewal of each central feature value less than predetermined threshold value, if so, then
Implementation center's feature vector generation module, if it is not, then returning to the first computing module;
Central feature vector generation module:For preserving each the central feature value after updating, a central feature vector is formed;
Central feature value determining module:For input feature vector matrix to be designated as into M1, for every a line therefrom heart characteristic vector of M1
In take out the central feature value of corresponding arest neighbors successively;
Characteristic vector output module:For counting the hit-count of each central feature value being removed and obtaining corresponding to described
The characteristic vector of input feature vector matrix.
Preferably, the characteristic vector in characteristic vector data storehouse is generated by with lower module:
Module one, for putting all pictures in picture database into queue in;
Module two, the local feature region for extracting every pictures in queue successively;
Module three, the eigenmatrix for calculating every pictures according to the local feature region per pictures;
Module four, it is converted to feature for the eigenmatrix of every pictures to be carried out into vectorization as input feature vector matrix successively
Vector, and corresponding characteristic vector will be preserved into characteristic vector data storehouse per pictures.
It will be apparent to those skilled in the art that technical scheme that can be as described above and design, make other various
It is corresponding to change and deformation, and all these change and deformation should all belong to the protection domain of the claims in the present invention
Within.
Claims (6)
1. a kind of picture retrieval matching process, it is characterised in that comprise the following steps:
S101, the picture to be matched for receiving user input;
S102, the local feature region for extracting picture to be matched;
S103, the eigenmatrix that picture to be matched is calculated according to the local feature region of picture to be matched;
S104, the eigenmatrix of picture to be matched is carried out into vectorization as input feature vector matrix it is converted to characteristic vector;
S105, the characteristic vector of picture to be matched is matched with the characteristic vector in characteristic vector data storehouse, and will matching
To characteristic vector corresponding to pictorial information return to user.
2. picture retrieval matching process according to claim 1, it is characterised in that the vectorization conversion includes following step
Suddenly:
A, the eigenmatrix for reading all pictures in picture database are simultaneously merged and obtain eigenmatrix M;
B, the random central feature value for generating predetermined number in the characteristic value space of eigenmatrix M;
C, the every a line for taking out eigenmatrix M, are designated as Fi, calculate successively and FiThe central feature value of arest neighbors;
D, each central feature value is updated successively, the rule of renewal is:It is special that current central feature value is updated to current center
The F of value indicative and arest neighborsiGeometric center value;
E, when judging that each central feature value updates produced displacement whether less than predetermined threshold value, if so, then perform step F,
If it is not, then return to step C;
Each central feature value after F, preservation renewal, forms a central feature vector;
G, input feature vector matrix is designated as M1, the every a line for M1 therefrom takes out corresponding arest neighbors in heart characteristic vector successively
Central feature value;
H, the hit-count for counting each central feature value being removed and obtain correspond to the input feature vector matrix feature
Vector.
3. picture retrieval matching process according to claim 2, it is characterised in that feature in characteristic vector data storehouse to
Amount is generated by following steps:
Step one into, all pictures in picture database are put queue;
Step 2, the local feature region for extracting every pictures in queue successively;
Step 3, the eigenmatrix that every pictures are calculated according to the local feature region per pictures;
Step 4, successively using the eigenmatrix of every pictures as input feature vector matrix carry out vectorization be converted to feature to
Amount, and corresponding characteristic vector will be preserved into characteristic vector data storehouse per pictures.
4. a kind of picture retrieval coalignment, it is characterised in that including with lower module:
Receiver module:Picture to be matched for receiving user input;
Feature point extraction module:Local feature region for extracting picture to be matched;
Eigenmatrix computing module:Feature square for calculating picture to be matched according to the local feature region of picture to be matched
Battle array;
Vectorization module:Spy is converted to for the eigenmatrix of picture to be matched to be carried out into vectorization as input feature vector matrix
Levy vector;
Matching returns to module:For the characteristic vector in the characteristic vector of picture to be matched and characteristic vector data storehouse to be carried out
Match somebody with somebody, and pictorial information corresponding to the characteristic vector that will match to returns to user.
5. picture retrieval coalignment according to claim 4, it is characterised in that the vectorization conversion includes following mould
Block:
Merging module:Eigenmatrix M is obtained for reading the eigenmatrix of all pictures in picture database and merging;
Central feature value generation module:Center for the random generation predetermined number in the characteristic value space of eigenmatrix M is special
Value indicative;
First computing module:Every a line for taking out eigenmatrix M, is designated as Fi, calculate successively and FiThe center of arest neighbors is special
Value indicative;
Central feature value update module:For updating each central feature value successively, the rule of renewal is:Current central feature
Value is updated to the F of current central feature value and arest neighborsiGeometric center value;
Judge module:For whether judging displacement produced during the renewal of each central feature value less than predetermined threshold value, if so, then
Implementation center's feature vector generation module, if it is not, then returning to the first computing module;
Central feature vector generation module:For preserving each the central feature value after updating, a central feature vector is formed;
Central feature value determining module:For input feature vector matrix to be designated as into M1, for every a line therefrom heart characteristic vector of M1
In take out the central feature value of corresponding arest neighbors successively;
Characteristic vector output module:For counting the hit-count of each central feature value being removed and obtaining corresponding to described
The characteristic vector of input feature vector matrix.
6. picture retrieval coalignment according to claim 5, it is characterised in that feature in characteristic vector data storehouse to
Amount is generated by with lower module:
Module one, for putting all pictures in picture database into queue in;
Module two, the local feature region for extracting every pictures in queue successively;
Module three, the eigenmatrix for calculating every pictures according to the local feature region per pictures;
Module four, it is converted to feature for the eigenmatrix of every pictures to be carried out into vectorization as input feature vector matrix successively
Vector, and corresponding characteristic vector will be preserved into characteristic vector data storehouse per pictures.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710068557.2A CN106844715A (en) | 2017-02-08 | 2017-02-08 | A kind of picture retrieval matching process and device |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710068557.2A CN106844715A (en) | 2017-02-08 | 2017-02-08 | A kind of picture retrieval matching process and device |
Publications (1)
Publication Number | Publication Date |
---|---|
CN106844715A true CN106844715A (en) | 2017-06-13 |
Family
ID=59122203
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710068557.2A Pending CN106844715A (en) | 2017-02-08 | 2017-02-08 | A kind of picture retrieval matching process and device |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106844715A (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111506751A (en) * | 2020-04-20 | 2020-08-07 | 创景未来(北京)科技有限公司 | Method and device for searching mechanical drawing |
US11663260B2 (en) | 2020-04-08 | 2023-05-30 | Baidu Online Network Technology (Beijing) Co., Ltd. | Method and apparatus for searching multimedia content device, and storage medium |
Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105718531A (en) * | 2016-01-14 | 2016-06-29 | 广州市万联信息科技有限公司 | Image database building method and image recognition method |
-
2017
- 2017-02-08 CN CN201710068557.2A patent/CN106844715A/en active Pending
Patent Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105718531A (en) * | 2016-01-14 | 2016-06-29 | 广州市万联信息科技有限公司 | Image database building method and image recognition method |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US11663260B2 (en) | 2020-04-08 | 2023-05-30 | Baidu Online Network Technology (Beijing) Co., Ltd. | Method and apparatus for searching multimedia content device, and storage medium |
CN111506751A (en) * | 2020-04-20 | 2020-08-07 | 创景未来(北京)科技有限公司 | Method and device for searching mechanical drawing |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US9348898B2 (en) | Recommendation system with dual collaborative filter usage matrix | |
US9454580B2 (en) | Recommendation system with metric transformation | |
CN102254015B (en) | Image retrieval method based on visual phrases | |
CN106815244B (en) | Text vector representation method and device | |
CN107688605B (en) | Cross-platform data matching process, device, computer equipment and storage medium | |
CN109816482B (en) | Knowledge graph construction method, device and equipment of e-commerce platform and storage medium | |
CN109829775A (en) | A kind of item recommendation method, device, equipment and readable storage medium storing program for executing | |
EP2829990A1 (en) | Image search device, image search method, program, and computer-readable storage medium | |
EP2833275B1 (en) | Image search device, image search method, program, and computer-readable storage medium | |
CN103164424A (en) | Method and device for acquiring time-efficient words | |
CN110390106B (en) | Semantic disambiguation method, device, equipment and storage medium based on two-way association | |
US20160098437A1 (en) | Information retrieval method and apparatus | |
CN112434188A (en) | Data integration method and device for heterogeneous database and storage medium | |
CN111125348A (en) | Text abstract extraction method and device | |
WO2015153240A1 (en) | Directed recommendations | |
CN106844715A (en) | A kind of picture retrieval matching process and device | |
CN102760127B (en) | Method, device and the equipment of resource type are determined based on expanded text information | |
CN109977286A (en) | Content-based information retrieval method | |
CN111339778B (en) | Text processing method, device, storage medium and processor | |
CN107665222B (en) | Keyword expansion method and device | |
TWI465949B (en) | Data clustering apparatus and method | |
CN109636509B (en) | Scoring prediction method for constructing submatrix based on asymmetric distance | |
CN106919712A (en) | The data statistical approach and system of form | |
CN107506572B (en) | Method and device for acquiring height of target point | |
CN112650869B (en) | Image retrieval reordering method and device, electronic equipment and storage medium |
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 | ||
WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20170613 |
|
WD01 | Invention patent application deemed withdrawn after publication |