CN106528662A - Quick retrieval method and system of vehicle image on the basis of feature geometric constraint - Google Patents
Quick retrieval method and system of vehicle image on the basis of feature geometric constraint Download PDFInfo
- Publication number
- CN106528662A CN106528662A CN201610916925.XA CN201610916925A CN106528662A CN 106528662 A CN106528662 A CN 106528662A CN 201610916925 A CN201610916925 A CN 201610916925A CN 106528662 A CN106528662 A CN 106528662A
- Authority
- CN
- China
- Prior art keywords
- hash
- vehicle image
- image
- picture
- 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
- 238000000034 method Methods 0.000 title claims abstract description 74
- 230000000007 visual effect Effects 0.000 claims abstract description 36
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 24
- 238000000605 extraction Methods 0.000 claims abstract description 20
- 238000004364 calculation method Methods 0.000 claims abstract description 11
- 239000013598 vector Substances 0.000 claims description 38
- 238000012800 visualization Methods 0.000 claims description 15
- 238000010276 construction Methods 0.000 claims description 8
- 238000013517 stratification Methods 0.000 claims description 7
- 238000003064 k means clustering Methods 0.000 claims description 6
- 238000012545 processing Methods 0.000 abstract description 2
- 238000005516 engineering process Methods 0.000 description 3
- 238000007689 inspection Methods 0.000 description 3
- 238000012544 monitoring process Methods 0.000 description 3
- 241000208340 Araliaceae Species 0.000 description 2
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 description 2
- 235000003140 Panax quinquefolius Nutrition 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 2
- 235000008434 ginseng Nutrition 0.000 description 2
- 230000004807 localization Effects 0.000 description 2
- 230000009466 transformation Effects 0.000 description 2
- PXFBZOLANLWPMH-UHFFFAOYSA-N 16-Epiaffinine Natural products C1C(C2=CC=CC=C2N2)=C2C(=O)CC2C(=CC)CN(C)C1C2CO PXFBZOLANLWPMH-UHFFFAOYSA-N 0.000 description 1
- 241001269238 Data Species 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000005286 illumination Methods 0.000 description 1
- 238000011835 investigation Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 238000012806 monitoring device Methods 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 230000007935 neutral effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
- 238000011524 similarity measure Methods 0.000 description 1
- 238000012549 training 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
- G06F16/58—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
- G06F16/583—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/46—Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features
- G06V10/462—Salient features, e.g. scale invariant feature transforms [SIFT]
-
- 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
- G06V10/755—Deformable models or variational models, e.g. snakes or active contours
- G06V10/7553—Deformable models or variational models, e.g. snakes or active contours based on shape, e.g. active shape models [ASM]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/08—Detecting or categorising vehicles
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Artificial Intelligence (AREA)
- Software Systems (AREA)
- Databases & Information Systems (AREA)
- Library & Information Science (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Probability & Statistics with Applications (AREA)
- Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- General Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Image Analysis (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention discloses a quick retrieval method and system of a vehicle image on the basis of a feature geometric constraint. The method comprises the following steps of: adopting a convolutional neural network to position the vehicle image, wherein the vehicle image comprises a query picture and a storage picture; carrying out feature extraction on the positioned image to obtain image features; adopting a hierarchical clustering method and a minimal hash method to carry out compressed encoding on the image features to obtain the hash code of the vehicle image features; according to the hash code of the vehicle image, adopting a visual phrase maintained by a geometric attribute to carry out image similarity calculation to obtain a quick matching retrieval result; and adopting a histogram of oriented gradients to carry out accurate matching on the quick matching retrieval result to obtain an accurate matching retrieval result. The method has the advantages of high retrieval accuracy and high retrieval speed, and can be widely applied to the field of image processing.
Description
Technical field
The present invention relates to image processing field, especially a kind of vehicle image quick-searching side of feature based geometrical constraint
Method and system.
Background technology
In recent years, with China's expanding economy, the increase of the trip demand of entire society so that number of vehicles is fast
Speed is soaring, and this brings many facilities to people's lives and work.However, while offering convenience, being disobeyed using automobile
Method criminal behavior relate to car case, robber robs vehicle case and various traffic violations are also more and more, threatens the people of resident
Body property safety.Such issues that in order to tackle, administration section, are additionally arranged various smart machines such as bayonet socket photographic head in city, survey
Fast instrument and fixed point sensor etc..Monitoring device and technology are developed rapidly, and substantial amounts of monitoring data is collected, and this is to pipe
Reason department brings great opportunities and challenges.If these data can be effectively utilized, it will provide pole to administration section's work
Big help.If but processed according still further to the method for traditional artificial investigation and screening, these data will become burden,
The manpower and materials of administration section can greatly be wasted.Therefore, how phase is quickly and efficiently retrieved from a large amount of monitoring datas
The useful information answered becomes an emphasis of future studies.
CBIR field, i.e. CBIR (Content-based image retrieval), are computers
In visual field with regard to large scale digital Image Retrieval research branch.Typical CBIR systems are based on content understanding side
Method, it is allowed to one pictures of user input, to search other pictures with same or similar content.Existing image search method
Mainly include two steps：Feature extraction and characteristic matching.Existing image search method first by or combination shallow-layer thing
Reason feature (such as color, texture, structure and shape etc.) carrys out phenogram picture, then by matching image to be retrieved and the spy of search library
Levy to weigh the similarity between them.Existing image search method by using or combination shallow-layer physical features such as face
Color, texture, structure and shape etc. carry out phenogram picture, it is impossible to effective expression image, semantic, and this will affect follow-up vehicle retrieval
Precision；Meanwhile, existing image search method directly carries out the matching of similarity using shallow-layer physical features, greatly can reduce
(because shallow-layer physical features are in order to retain image information as far as possible, its quantity and dimension can excessively redundancies for follow-up retrieval rate.
Therefore, need in the industry a kind of image that effective information quickly and accurately can be filtered out from magnanimity monitoring data badly
Search method and system.
The content of the invention
To solve above-mentioned technical problem, it is an object of the invention to：There is provided a kind of retrieval precision high fast with retrieval rate,
The vehicle image method for quickly retrieving of feature based geometrical constraint.
Another object of the present invention is to：There is provided a kind of retrieval precision high fast with retrieval rate, feature based geometry is about
The vehicle image quick retrieval system of beam.
The technical solution used in the present invention is：
A kind of vehicle image method for quickly retrieving of feature based geometrical constraint, comprises the following steps：
Vehicle image is positioned using convolutional neural networks, the vehicle image includes inquiring about picture and warehouse-in figure
Piece；
Image to orienting carries out feature extraction, obtains characteristics of image；
Coding is compressed using the method for hierarchical clustering and min-hash to characteristics of image, vehicle image feature is obtained
Hash coding；
Encoded using the visualization phrase of geometric attribute holding to carry out image similarity meter according to the Hash of vehicle image
Calculate, obtain Rapid matching retrieval result；
Accurately mate is carried out to Rapid matching retrieval result using histograms of oriented gradients, exact match search knot is obtained
Really.
Further, the convolutional neural networks adopt AlexNet convolutional neural networks, the AlexNet convolutional Neurals net
Network includes 5 convolutional layers and 3 full articulamentums, and the output layer of the AlexNet convolutional neural networks is four dimensional vectors, should
Four dimensional vectors represent four coordinates of target boundary rectangle.
Further, the described pair of image oriented carries out feature extraction, and the step for obtain characteristics of image, which is specially：
Feature extraction is carried out to the image oriented using Scale invariant features transform, vehicle image A is obtainedISIFT it is special
X, the vehicle image A are closed in collectionIThe expression formula of SIFT feature set X be：X=[x1,x2,...xN]∈RD×N, wherein, N is
The total quantity of SIFT feature point, xiFor the characteristic vector of i-th SIFT feature point, i=1,2 ... ... N, D are characterized the dimension of vector
Number, RD×NRepresent D × N-dimensional vector space.
Further, the method for the employing hierarchical clustering and min-hash is compressed coding to SIFT feature, obtains
The step for Hash of vehicle image feature is encoded, which includes：
Using Hierarchical K-Means hierarchical clusterings algorithms to vehicle image AISIFT feature carry out dictionary
Cluster, obtains vehicle image AIVisual vocabulary set W, the vehicle image AIThe expression formula of visual vocabulary set W be：W=
[w1,w2,...wN]∈RN, wherein, RNRepresent N-dimensional vector space, wiIt is for the visual vocabulary of i-th SIFT feature point, described
Hierarchical K-Means hierarchical clustering methods are the stratification K-Means clustering algorithms of tree construction：First, all units
Element regards a cluster as, and in the ground floor of tree, all data points is divided into and is wrapped for final visual dictionary size
Containing the less cluster centre of cluster, the number K=10 of per layer of cluster centre；Then, K-Means side is adopted to each cluster centre
Method carries out cluster calculation, is obtained by 10TThe visual vocabulary of individual leaf node composition, T represent the number of plies of tree；
According to vehicle image AIVisual vocabulary set W Hash coding is carried out using the method for min-hash, obtain vehicle
Image AIMinimum hash mfj(AI), the method for the min-hash first generates the hash function of N number of independent random, Ran Hougen
According to vehicle image AIVisual vocabulary set W calculate corresponding cryptographic Hash, be worth to vehicle finally according to the Hash for calculating
Image AIMinimum hash mfj(AI), minimum hash mfj(AI) expression formula be：mfj(AI)=argminw∈Wfj
(w), wherein, fjFor j-th hash function in N number of hash function, w is image AIArbitrary visual vocabulary.
Further, it is described to be encoded using the visualization phrase of geometric attribute holding to carry out figure according to the Hash of vehicle image
As Similarity Measure, the step for obtain Rapid matching retrieval result, which includes：
According to vehicle image AIMinimum hash mfj(AI) obtain inquiry picture A1Minimum hash and warehouse-in picture A2Most
The position of collision of little cryptographic Hash；
An offset space is set up, the offset space divides the space into M × M region, and M is the regulation ginseng of setting
Number；
Will inquiry picture A1Minimum hash and warehouse-in picture A2Position of collision be mapped in offset space, and statistics look into
Ask picture A1Minimum hash and warehouse-in picture A2Collision frequency of the minimum hash in each region in offset space；
It is inquiry picture A according to the collision frequency of statistics1With warehouse-in picture A2Similarity given a mark；
Inquiry picture A is drawn according to the result of marking1With warehouse-in picture A2Rapid matching retrieval result.
Further, the employing histograms of oriented gradients carries out accurately mate to Rapid matching retrieval result, obtains accurate
The step for matching retrieval result, which includes：
The feature of user's area-of-interest is extracted using histograms of oriented gradients from Rapid matching retrieval result；
Calculated in the distance of the adjacent position travel direction histogram of gradients of picture to be matched by the scanning window for setting
Compare, so as to obtain exact match search result.
Another technical scheme for being taken of the present invention is：
A kind of vehicle image quick retrieval system of feature based geometrical constraint, including with lower module：
Locating module, for being positioned to vehicle image using convolutional neural networks, the vehicle image includes inquiry
Picture and warehouse-in picture；
Characteristic extracting module, for carrying out feature extraction to the image oriented using Scale invariant features transform, obtains
SIFT feature；
Compressed encoding module, for volume is compressed to SIFT feature using the method for hierarchical clustering and min-hash
Code, obtains the Hash coding of vehicle image；
Rapid matching module, for according to the Hash of vehicle image coding using geometric attribute keep visualization phrase come
Image similarity calculating is carried out, Rapid matching retrieval result is obtained；
Accurately mate module, for carrying out accurately mate to Rapid matching retrieval result using histograms of oriented gradients, obtains
To exact match search result.
Further, the characteristic extracting module is performed and is operated in detail below：
Feature extraction is carried out to the image oriented using Scale invariant features transform, vehicle image A is obtainedISIFT it is special
X, the vehicle image A are closed in collectionIThe expression formula of SIFT feature set X be：X=[x1,x2,...xN]∈RD×N, wherein, N is
The total quantity of SIFT feature point, xiFor the characteristic vector of i-th SIFT feature point, i=1,2 ... ... N, D are characterized the dimension of vector
Number, RD×NRepresent D × N-dimensional vector space.
Further, the compressed encoding module includes：
Dictionary cluster cell, for using Hierarchical K-Means hierarchical clusterings algorithms to vehicle image AI's
SIFT feature carries out dictionary cluster, obtains vehicle image AIVisual vocabulary set W, the vehicle image AIVisual word collect
Close W expression formula be：W=[w1,w2,...wN]∈RN, wherein, RNRepresent N-dimensional vector space, wiFor i-th SIFT feature point
Visual vocabulary, the Hierarchical K-Means hierarchical clustering methods are the stratification K-Means clustering algorithms of tree construction：
First, all elements are regarded as a cluster, and in the ground floor of tree, all data points is divided into relative to final visual word
Comprising the cluster centre that cluster is less, the number K=10 of per layer of cluster centre for allusion quotation size；Then, to each cluster centre
Cluster calculation is carried out using K-Means methods, is obtained by 10TThe visual vocabulary of individual leaf node composition, T represent the layer of tree
Number；
Hash coding unit, for according to vehicle image AIVisual vocabulary set W carried out using the method for min-hash
Hash is encoded, and obtains vehicle image AIMinimum hash mfj(AI), the method for the min-hash first generates N number of independent random
Hash function, then according to vehicle image AIVisual vocabulary set W calculate corresponding cryptographic Hash, finally according to calculating
Hash be worth to vehicle image AIMinimum hash mfj(AI), minimum hash mfj(AI) expression formula be：mfj
(AI)=argminw∈Wfj(w), wherein, fjFor j-th hash function in N number of hash function, w is image AIArbitrary vision
Vocabulary.
Further, the Rapid matching module includes：
Position of collision acquiring unit, for according to vehicle image AIMinimum hash mfj(AI) obtain inquiry picture A1Most
Little cryptographic Hash and warehouse-in picture A2The position of collision of minimum hash；
Offset space sets up unit, and for setting up an offset space, the offset space divides the space into M × M
Region, M are the regulation parameter of setting；
Map unit, for picture A will be inquired about1Minimum hash and warehouse-in picture A2The position of collision of minimum hash reflects
It is mapped in offset space, and statistical query picture A1Minimum hash and warehouse-in picture A2Minimum hash is every in offset space
Collision frequency in individual region；
Marking unit, is inquiry picture A for the collision frequency according to statistics1With warehouse-in picture A2Similarity carry out beat
Point；
Matching unit, for drawing inquiry picture A according to the result of marking1With warehouse-in picture A2Rapid matching retrieval knot
Really.
The beneficial effects of the method for the present invention is：Vehicle image is positioned using convolutional neural networks, and adopt layer
Secondaryization is clustered and the method for min-hash is compressed coding to characteristics of image, is reduced redundancy, is improve follow-up inspection
Suo Sudu；Encoded using the visualization phrase of geometric attribute holding to carry out image similarity meter according to the Hash of vehicle image
Calculate, construct the vehicle image characteristic manner with geometrical constraint characteristic, protected by the visualization phrase of geometric attribute holding
The spatial information of image is stayed, the precision of retrieval is improve；Essence is carried out to Rapid matching retrieval result using histograms of oriented gradients
Really match, Secondary Match is carried out to vehicle by histograms of oriented gradients, the precision of retrieval is improve.
The beneficial effect of system of the present invention is：It is fixed vehicle image to be carried out using convolutional neural networks in locating module
Position, and in compressed encoding module, adopt hierarchical clustering and the method for min-hash that coding is compressed to characteristics of image, subtract
Lack redundancy, improve follow-up retrieval rate；Adopted according to the Hash of vehicle image coding in Rapid matching module
The visualization phrase that geometric attribute keeps constructs the vehicle image with geometrical constraint characteristic carrying out image similarity calculating
Characteristic manner, retains the spatial information of image by the visualization phrase of geometric attribute holding, improves the precision of retrieval；
Accurately mate is carried out using histograms of oriented gradients to Rapid matching retrieval result in accurately mate module, it is straight by direction gradient
Square figure improves the precision of retrieval carrying out Secondary Match to vehicle.
Description of the drawings
Fig. 1 is a kind of overall flow figure of the vehicle image method for quickly retrieving of feature based geometrical constraint of the invention；
Fig. 2 is the process schematic of vehicle location of the present invention；
Fig. 3 is the process schematic that SIFT feature of the present invention is extracted；
Fig. 4 is the structural representation of Hierarchical K-Means dictionaries of the present invention；
Fig. 5 is the process schematic that the present invention calculates similar score in offset space.
Specific embodiment
With reference to Fig. 1, a kind of vehicle image method for quickly retrieving of feature based geometrical constraint is comprised the following steps：
Vehicle image is positioned using convolutional neural networks, the vehicle image includes inquiring about picture and warehouse-in figure
Piece；
Image to orienting carries out feature extraction, obtains characteristics of image；
Coding is compressed using the method for hierarchical clustering and min-hash to characteristics of image, vehicle image feature is obtained
Hash coding；
Encoded using the visualization phrase of geometric attribute holding to carry out image similarity meter according to the Hash of vehicle image
Calculate, obtain Rapid matching retrieval result；
Accurately mate is carried out to Rapid matching retrieval result using histograms of oriented gradients, exact match search knot is obtained
Really.
It is further used as preferred embodiment, the convolutional neural networks adopt AlexNet convolutional neural networks, described
AlexNet convolutional neural networks include 5 convolutional layers and 3 full articulamentums, the output layer of the AlexNet convolutional neural networks
For four dimensional vectors, four dimensional vector represents four coordinates of target boundary rectangle.
It is further used as preferred embodiment, the described pair of image oriented carries out feature extraction, obtains characteristics of image
The step for, which is specially：
Feature extraction is carried out to the image oriented using Scale invariant features transform, vehicle image A is obtainedISIFT it is special
X, the vehicle image A are closed in collectionIThe expression formula of SIFT feature set X be：X=[x1,x2,...xN]∈RD×N, wherein, N is
The total quantity of SIFT feature point, xiFor the characteristic vector of i-th SIFT feature point, i=1,2 ... ... N, D are characterized the dimension of vector
Number, RD×NRepresent D × N-dimensional vector space.
It is further used as preferred embodiment, the method for the employing hierarchical clustering and min-hash is to SIFT feature
Coding is compressed, the step for Hash of vehicle image feature is encoded is obtained, which includes：
Using Hierarchical K-Means hierarchical clusterings algorithms to vehicle image AISIFT feature carry out dictionary
Cluster, obtains vehicle image AIVisual vocabulary set W, the vehicle image AIThe expression formula of visual vocabulary set W be：W=
[w1,w2,...wN]∈RN, wherein, RNRepresent N-dimensional vector space, wiIt is for the visual vocabulary of i-th SIFT feature point, described
Hierarchical K-Means hierarchical clustering methods are the stratification K-Means clustering algorithms of tree construction：First, all units
Element regards a cluster as, and in the ground floor of tree, all data points is divided into and is wrapped for final visual dictionary size
Containing the less cluster centre of cluster, the number K=10 of per layer of cluster centre；Then, K-Means side is adopted to each cluster centre
Method carries out cluster calculation, is obtained by 10TThe visual vocabulary of individual leaf node composition, T represent the number of plies of tree；
According to vehicle image AIVisual vocabulary set W Hash coding is carried out using the method for min-hash, obtain vehicle
Image AIMinimum hash mfj(AI), the method for the min-hash first generates the hash function of N number of independent random, Ran Hougen
According to vehicle image AIVisual vocabulary set W calculate corresponding cryptographic Hash, be worth to vehicle finally according to the Hash for calculating
Image AIMinimum hash mfj(AI), minimum hash mfj(AI) expression formula be：mfj(AI)=argminw∈Wfj
(w), wherein, fjFor j-th hash function in N number of hash function, w is image AIArbitrary visual vocabulary.
It is further used as preferred embodiment, it is described to adopt what geometric attribute kept according to the Hash of vehicle image coding
Carrying out image similarity calculating, the step for obtain Rapid matching retrieval result, which includes visualization phrase：
According to vehicle image AIMinimum hash mfj(AI) obtain inquiry picture A1Minimum hash and warehouse-in picture A2Most
The position of collision of little cryptographic Hash；
An offset space is set up, the offset space divides the space into M × M region, and M is the regulation ginseng of setting
Number；
Will inquiry picture A1Minimum hash and warehouse-in picture A2Position of collision be mapped in offset space, and statistics look into
Ask picture A1Minimum hash and warehouse-in picture A2Collision frequency of the minimum hash in each region in offset space；
It is inquiry picture A according to the collision frequency of statistics1With warehouse-in picture A2Similarity given a mark；
Inquiry picture A is drawn according to the result of marking1With warehouse-in picture A2Rapid matching retrieval result.
It is further used as preferred embodiment, the employing histograms of oriented gradients is carried out to Rapid matching retrieval result
Accurately mate, the step for obtain exact match search result, which includes：
The feature of user's area-of-interest is extracted using histograms of oriented gradients from Rapid matching retrieval result；
Calculated in the distance of the adjacent position travel direction histogram of gradients of picture to be matched by the scanning window for setting
Compare, so as to obtain exact match search result.
With reference to Fig. 1, a kind of vehicle image quick retrieval system of feature based geometrical constraint, including with lower module：
Locating module, for being positioned to vehicle image using convolutional neural networks, the vehicle image includes inquiry
Picture and warehouse-in picture；
Characteristic extracting module, for carrying out feature extraction to the image oriented using Scale invariant features transform, obtains
SIFT feature；
Compressed encoding module, for volume is compressed to SIFT feature using the method for hierarchical clustering and min-hash
Code, obtains the Hash coding of vehicle image；
Rapid matching module, for according to the Hash of vehicle image coding using geometric attribute keep visualization phrase come
Image similarity calculating is carried out, Rapid matching retrieval result is obtained；
Accurately mate module, for carrying out accurately mate to Rapid matching retrieval result using histograms of oriented gradients, obtains
To exact match search result.
It is further used as preferred embodiment, the characteristic extracting module is performed and operated in detail below：
Feature extraction is carried out to the image oriented using Scale invariant features transform, vehicle image A is obtainedISIFT it is special
X, the vehicle image A are closed in collectionIThe expression formula of SIFT feature set X be：X=[x1,x2,...xN]∈RD×N, wherein, N is
The total quantity of SIFT feature point, xiFor the characteristic vector of i-th SIFT feature point, i=1,2 ... ... N, D are characterized the dimension of vector
Number, RD×NRepresent D × N-dimensional vector space.
It is further used as preferred embodiment, the compressed encoding module includes：
Dictionary cluster cell, for using Hierarchical K-Means hierarchical clusterings algorithms to vehicle image AI's
SIFT feature carries out dictionary cluster, obtains vehicle image AIVisual vocabulary set W, the vehicle image AIVisual word collect
Close W expression formula be：W=[w1,w2,...wN]∈RN, wherein, RNRepresent N-dimensional vector space, wiFor i-th SIFT feature point
Visual vocabulary, the Hierarchical K-Means hierarchical clustering methods are the stratification K-Means clustering algorithms of tree construction：
First, all elements are regarded as a cluster, and in the ground floor of tree, all data points is divided into relative to final visual word
Comprising the cluster centre that cluster is less, the number K=10 of per layer of cluster centre for allusion quotation size；Then, to each cluster centre
Cluster calculation is carried out using K-Means methods, is obtained by 10TThe visual vocabulary of individual leaf node composition, T represent the layer of tree
Number；
Hash coding unit, for according to vehicle image AIVisual vocabulary set W carried out using the method for min-hash
Hash is encoded, and obtains vehicle image AIMinimum hash mfj(AI), the method for the min-hash first generates N number of independent random
Hash function, then according to vehicle image AIVisual vocabulary set W calculate corresponding cryptographic Hash, finally according to calculating
Hash be worth to vehicle image AIMinimum hash mfj(AI), minimum hash mfj(AI) expression formula be：mfj
(AI)=argminw∈Wfj(w), wherein, fjFor j-th hash function in N number of hash function, w is image AIArbitrary vision
Vocabulary.
Wherein, for final visual dictionary size refer to and compare last comprising the cluster centre that cluster is less
For the cluster sum for being included of layer, comprising the less layer of cluster, (for example, last layer may have 10000 clusters, but ground floor
Cluster may only have 10).
It is further used as preferred embodiment, the Rapid matching module includes：
Position of collision acquiring unit, for according to vehicle image AIMinimum hash mfj(AI) obtain inquiry picture A1Most
Little cryptographic Hash and warehouse-in picture A2The position of collision of minimum hash；
Offset space sets up unit, and for setting up an offset space, the offset space divides the space into M × M
Region, M are the regulation parameter of setting；
Map unit, for picture A will be inquired about1Minimum hash and warehouse-in picture A2The position of collision of minimum hash reflects
It is mapped in offset space, and statistical query picture A1Minimum hash and warehouse-in picture A2Minimum hash is every in offset space
Collision frequency in individual region；
Marking unit, is inquiry picture A for the collision frequency according to statistics1With warehouse-in picture A2Similarity carry out beat
Point；
Matching unit, for drawing inquiry picture A according to the result of marking1With warehouse-in picture A2Rapid matching retrieval knot
Really.
The present invention is described in further detail with reference to Figure of description and specific embodiment.
Embodiment one
For prior art retrieval precision is low and the slow problem of retrieval rate, the present invention proposes a kind of feature based geometry
The vehicle image method for quickly retrieving of constraint and system.The system can be divided into two parts according to function：A part is quick
Matching feature, another part are accurately mate functions.In Rapid matching function, convolution of the system first with current forward position
Neutral net reduces redundancy, improves follow-up retrieval precision and speed positioning to vehicle pictures；Then it is sharp
Characteristics of image is extracted with Scale invariant features transform (Scale-invariant feature transform, SIFT), is then led to
The method for crossing hierarchical clustering (Hierarchical K-Means, HKM) and min-hash (Min-Hash) is come to characteristics of image
Coding is compressed, the effect of quick-searching is reached；It is last to carry out adding a kind of geometric attribute when image similarity is calculated
The visualization phrase (Geometry-Preserving Visual Phrases, GVP) of holding retaining the spatial information of image,
Improve the precision of retrieval.And in accurately mate function, system utilization orientation histogram of gradients (Histogram of
Oriented Gradient, HOG) to carrying out stencil matching near user-defined area-of-interest, realizing self-defined height
The quadratic search of precision.
As shown in figure 1, the present invention is by input inquiry vehicle pictures (i.e. suspicious picture) C, export similar to suspect vehicle
All vehicle pictures, and carry out quicksort according to retrieving similarity.The system of the present invention consists predominantly of 6 modules, point
Be not vehicle localization module, characteristic extracting module, dictionary cluster module, Hash coding module, GVP similarity calculation modules and
Accurately mate module.The whole system of the present invention includes offline and online the two patterns, and in off-line mode, system passes through
Vehicle localization module, characteristic extracting module, dictionary cluster module, Hash coding module are warehouse-in picture (by picture y1, y2... ...
ynComposition) (sign y to be converted to Hash coding1, y2... ... yn) be stored in data base；And in line model, system is first
Inquiry picture C is converted to Hash coding (sign C), then inquiry picture is realized and is entered by GVP similarity calculation modules
The quick-searching of storehouse picture, finally needs further according to actual service inquiry, carries out secondary accurate inspection using accurately mate module
Rope.
As shown in figure 1, the detailed process of the vehicle image method for quickly retrieving of feature based geometrical constraint of the present invention is as follows：
(1) vehicle location.
Vehicle pictures generally comprise more background, and region interested only includes the region of vehicle, so being
Reduce redundancy and improve follow-up retrieval precision, system first with current forward position convolutional neural networks to vehicle figure
Piece is positioned.Convolutional neural networks framework of the present invention based on vehicle location is proposed in 2012 with reference to Krizhevsky et al.
AlexNet convolutional neural networks, have 5 convolutional layers and 3 full articulamentums, its output layer is four dimensional vectors, represent
Four coordinates (X1, Y1, X2, Y2) of target boundary rectangle, as shown in Figure 2.Wherein, the implication of 55 × 55 × 96 in Fig. 2
That convolution is carried out to input picture using 96 convolution kernels, obtain 96 it is big it is little be 55 × 55 characteristic pattern, other parameters implication
Also it is similar to.
(2) feature extraction.
The present invention is right using Scale invariant features transform (Scale-invariant feature transform, SIFT)
The picture oriented carries out feature extraction, obtains characterizing the characteristic vector of information of vehicles.The result that feature of present invention is extracted is such as
Shown in Fig. 3, the vector set that piece image mapping transformation is a local feature, each characteristic vector are had flat by SIFT feature
The invariance moved, rotate, scale, while also there is certain invariance to illumination variation, affine and projective transformation, so with SIFT
Feature to extract vehicle pictures just can effectively retain the information of vehicles of picture.
After feature extraction, vehicle image AIJust can be represented with the set X of one group of SIFT feature：
X=[x1,x2,...xN]∈RD×N (1)
Wherein, N is the total quantity of image characteristic point, xiFor the characteristic vector of ith feature point, D is the dimension of characteristic vector
Number, the D optional 128 of the present embodiment.
(3) hierarchical clustering.
After extracting image characteristic point, it is possible to weigh it by calculating the distance between characteristic vector of two width images
Similarity, but the image of 500 × 500 will contain 750 SIFT feature points, directly comparison characteristic vector meter
Calculation amount is larger, it will have a strong impact on the search efficiency of system, so needing to reduce the characteristic point for being used for characterizing picture.The present invention is adopted
With the method for Hierarchical K-Means, SIFT feature point visual word is remitted and is represented, this advantageously reduces does not have
The SIFT feature point of obvious characteristic and the impact of picture noise, at utmost remain with while characteristic dimension is reduced
Information.
Hierarchical K-Means are the stratification K-Means of tree construction, and it regards one as all elements first
Cluster, in the ground floor of tree, all data points are all divided into the less cluster centre of cluster (relative to final visual dictionary size
For), per layer of cluster centre is K=10；Then, each cluster centre is calculated using K-Means methods, last has
Obtain 10TIndividual leaf node, that is, have 10TIndividual visual vocabulary, what T was represented are the numbers of plies of tree, and this tree construction is exactly that training in advance is good
Dictionary, child node is exactly visual vocabulary, as shown in Figure 4.Compared with traditional K-Means, Hierarchical K-Means
Tree structure can quickly and correctly to inquire about characteristic point classify.After the completion of dictionary cluster, image AIJust can be with
Represented with the set W of visual vocabulary：
W=[w1,w2,...wN]∈RN (2)
(4) Hash coding.
In order to continue to lift up recall precision, the system of the present invention introduces the method for min-hash (Min-Hash) to represent
Image.Min-Hash be it is a kind of it is quick judge two set whether similar technology, it firstly generates the Hash of N number of independent random
Function, then each hash function corresponding cryptographic Hash can be calculated according to visual vocabulary.Therefore image AIUnder hash function j
Minimum hash be defined as：
mfj(AI)=argminw∈Wfj(w) (3)
Wherein, fjFor j-th hash function in N number of hash function, w is image AIArbitrary visual vocabulary.
It is calculated two image AIAnd ATAfter the minimum hash under J hash function, the similarity of two set can
It is calculated with by collision (the possessing identical minimum hash) probability under identical hash function：
Assume that L is two image AIAnd ATThe number of times collided in J hash function, then their similarity is equal to
L/J。
(5) GVP image similarities are calculated.
The present invention introduces the visualization phrase (Geometry- that a kind of geometric attribute keeps when image similarity is calculated
Preserving Visual Phrases, GVP) improving retrieval precision.The GVP image similarity calculation methods of the present invention are first
An offset space as shown in Figure 5 is first set up, space is divided into M × M region, wherein, M is adjustable parameter；Then
The position of the minimum hash collision of two images is mapped to this offset space, and counts in interior each region of the offset space
Collision frequency, the collision frequency of the same area is higher, then its similar score is higher.This method remains the sky of vehicle image
Between structural constraint, eliminate the interference of noise, be effectively improved the degree of accuracy of retrieval.
(6) accurately mate.
For the inspection of user's some local features interested (such as the ornament of vehicle window, car light of depression etc.) to vehicle pictures
Rope demand, the system add accurately mate functional module.The accurately mate module of the present invention is first with direction gradient Nogata
Scheme (Histogram of Oriented Gradient, HOG) to extract the feature of user's area-of-interest, followed by scanning
Window carries out the histogrammic distances of HOG in the adjacent position of picture to be matched and calculates comparison, so as to obtain the result of accurately mate.
Compared with prior art, the present invention has advantages below：
1) for vehicle pictures the characteristics of, the quick and precisely positioning of vehicle is realized using convolutional neural networks, reduced
Redundancy, improves later retrieval speed.
2) using the method for Hierarchical K-Means clusters, min-hash and GVP, construct efficiently and accurately simultaneously
Vehicle pictures expression way with geometrical constraint characteristic, improves vehicle pictures warehouse-in and the speed retrieved and precision.
3) there is the region (i.e. user-defined area-of-interest) of obvious characteristic to carry out feature using HOG to carry to vehicle
Take and stencil matching, the precision of retrieval is improve by way of quadratic search.
It is more than that the preferable enforcement to the present invention is illustrated, but the present invention is not limited to the embodiment, it is ripe
Know those skilled in the art a variety of equivalent variations or replacement can be also made on the premise of without prejudice to spirit of the invention, this
The deformation or replacement of a little equivalents is all contained in the application claim limited range.
Claims (10)
1. a kind of vehicle image method for quickly retrieving of feature based geometrical constraint, it is characterised in that：Comprise the following steps：
Vehicle image is positioned using convolutional neural networks, the vehicle image includes inquiring about picture and warehouse-in picture；
Image to orienting carries out feature extraction, obtains characteristics of image；
Coding is compressed using the method for hierarchical clustering and min-hash to characteristics of image, the Kazakhstan of vehicle image feature is obtained
Uncommon coding；
Encoded using the visualization phrase of geometric attribute holding to carry out image similarity calculating according to the Hash of vehicle image, obtained
To Rapid matching retrieval result；
Accurately mate is carried out to Rapid matching retrieval result using histograms of oriented gradients, exact match search result is obtained.
2. the vehicle image method for quickly retrieving of a kind of feature based geometrical constraint according to claim 1, its feature exist
In：The convolutional neural networks adopt AlexNet convolutional neural networks, the AlexNet convolutional neural networks to include 5 convolution
Layer and 3 full articulamentums, the output layer of the AlexNet convolutional neural networks is four dimensional vectors, four dimensional vectors representative
Four coordinates of target boundary rectangle.
3. the vehicle image method for quickly retrieving of a kind of feature based geometrical constraint according to claim 1, its feature exist
In：The described pair of image oriented carries out feature extraction, and the step for obtain characteristics of image, which is specially：
Feature extraction is carried out to the image oriented using Scale invariant features transform, vehicle image A is obtainedISIFT feature collection
Close X, the vehicle image AIThe expression formula of SIFT feature set X be：X=[x1,x2,...xN]∈RD×N, wherein, N is SIFT
The total quantity of characteristic point, xiFor the characteristic vector of i-th SIFT feature point, i=1,2 ... ... N, D are characterized the dimension of vector, RD ×NRepresent D × N-dimensional vector space.
4. the vehicle image method for quickly retrieving of a kind of feature based geometrical constraint according to claim 3, its feature exist
In：The method of the employing hierarchical clustering and min-hash is compressed coding to SIFT feature, obtains vehicle image feature
Hash the step for encode, which includes：
Using Hierarchical K-Means hierarchical clusterings algorithms to vehicle image AISIFT feature carry out dictionary cluster,
Obtain vehicle image AIVisual vocabulary set W, the vehicle image AIThe expression formula of visual vocabulary set W be：W=[w1,
w2,...wN]∈RN, wherein, RNN-dimensional vector space is represented, wi is the visual vocabulary of i-th SIFT feature point, described
Hierarchical K-Means hierarchical clustering methods are the stratification K-Means clustering algorithms of tree construction：First, all units
Element regards a cluster as, and in the ground floor of tree, all data points is divided into and is wrapped for final visual dictionary size
Containing the less cluster centre of cluster, the number K=10 of per layer of cluster centre；Then, K-Means side is adopted to each cluster centre
Method carries out cluster calculation, is obtained by 10TThe visual vocabulary of individual leaf node composition, T represent the number of plies of tree；
According to vehicle image AIVisual vocabulary set W Hash coding is carried out using the method for min-hash, obtain vehicle image AI
Minimum hash mfj(AI), the method for the min-hash first generates the hash function of N number of independent random, then according to vehicle
Image AIVisual vocabulary set W calculate corresponding cryptographic Hash, be worth to vehicle image A finally according to the Hash for calculatingI's
Minimum hash mfj(AI), minimum hash mfj(AI) expression formula be：mfj(AI)=arg minw∈Wfj(w), wherein,
fjFor j-th hash function in N number of hash function, w is image AIArbitrary visual vocabulary.
5. the vehicle image method for quickly retrieving of a kind of feature based geometrical constraint according to claim 4, its feature exist
In：It is described to be encoded using the visualization phrase of geometric attribute holding to carry out image similarity meter according to the Hash of vehicle image
Calculate, the step for obtain Rapid matching retrieval result, which includes：
According to vehicle image AIMinimum hash mfj(AI) obtain inquiry picture A1Minimum hash and warehouse-in picture A2It is minimum to breathe out
The position of collision of uncommon value；
An offset space is set up, the offset space divides the space into M × M region, and M is the regulation parameter of setting；
Will inquiry picture A1Minimum hash and warehouse-in picture A2Position of collision be mapped in offset space, and statistical query figure
Piece A1Minimum hash and warehouse-in picture A2Collision frequency of the minimum hash in each region in offset space；
It is inquiry picture A according to the collision frequency of statistics1With warehouse-in picture A2Similarity given a mark；
Inquiry picture A is drawn according to the result of marking1With warehouse-in picture A2Rapid matching retrieval result.
6. the vehicle image method for quickly retrieving of a kind of feature based geometrical constraint according to any one of claim 1-5,
It is characterized in that：The employing histograms of oriented gradients carries out accurately mate to Rapid matching retrieval result, obtains accurately mate
The step for retrieval result, which includes：
The feature of user's area-of-interest is extracted using histograms of oriented gradients from Rapid matching retrieval result；
Calculated in the distance of the adjacent position travel direction histogram of gradients of picture to be matched by the scanning window for setting and compared,
So as to obtain exact match search result.
7. a kind of vehicle image quick retrieval system of feature based geometrical constraint, it is characterised in that：Including with lower module：
Locating module, for being positioned to vehicle image using convolutional neural networks, the vehicle image includes inquiring about picture
With warehouse-in picture；
Characteristic extracting module, for carrying out feature extraction to the image oriented using Scale invariant features transform, obtains SIFT
Feature；
Compressed encoding module, for coding is compressed to SIFT feature using the method for hierarchical clustering and min-hash, obtains
Hash to vehicle image is encoded；
Rapid matching module, is carried out for being encoded according to the Hash of vehicle image using the visualization phrase of geometric attribute holding
Image similarity is calculated, and obtains Rapid matching retrieval result；
Accurately mate module, for carrying out accurately mate to Rapid matching retrieval result using histograms of oriented gradients, obtains essence
Retrieval result is matched really.
8. the vehicle image quick retrieval system of a kind of feature based geometrical constraint according to claim 7, its feature exist
In：The characteristic extracting module is performed and is operated in detail below：
Feature extraction is carried out to the image oriented using Scale invariant features transform, vehicle image A is obtainedISIFT feature collection
Close X, the vehicle image AIThe expression formula of SIFT feature set X be：X=[x1,x2,...xN]∈RD×N, wherein, N is SIFT
The total quantity of characteristic point, xiFor the characteristic vector of i-th SIFT feature point, i=1,2 ... ... N, D are characterized the dimension of vector, RD ×NRepresent D × N-dimensional vector space.
9. the vehicle image quick retrieval system of a kind of feature based geometrical constraint according to claim 8, its feature exist
In：The compressed encoding module includes：
Dictionary cluster cell, for using Hierarchical K-Means hierarchical clusterings algorithms to vehicle image AISIFT
Feature carries out dictionary cluster, obtains vehicle image AIVisual vocabulary set W, the vehicle image AIVisual vocabulary set W
Expression formula is：W=[w1,w2,...wN]∈RN, wherein, RNRepresent N-dimensional vector space, wiFor the vision of i-th SIFT feature point
Vocabulary, the Hierarchical K-Means hierarchical clustering methods are the stratification K-Means clustering algorithms of tree construction：It is first
First, all elements are regarded as a cluster, and in the ground floor of tree, all data points is divided into relative to final visual dictionary
Comprising the cluster centre that cluster is less, the number K=10 of per layer of cluster centre for size；Then, each cluster centre is adopted
Cluster calculation is carried out with K-Means methods, is obtained by 10TThe visual vocabulary of individual leaf node composition, T represent the number of plies of tree；
Hash coding unit, for according to vehicle image AIVisual vocabulary set W Hash volume is carried out using the method for min-hash
Code, obtains vehicle image AIMinimum hash mfj(AI), the method for the min-hash first generates the Hash of N number of independent random
Function, then according to vehicle image AIVisual vocabulary set W calculate corresponding cryptographic Hash, finally according to the Hash for calculating
It is worth to vehicle image AIMinimum hash mfj(AI), minimum hash mfj(AI) expression formula be：mfj(AI)=
argminw∈W fj(w), wherein, fjFor j-th hash function in N number of hash function, w is image AIArbitrary visual vocabulary.
10. the vehicle image quick retrieval system of a kind of feature based geometrical constraint according to claim 9, its feature exist
In：The Rapid matching module includes：
Position of collision acquiring unit, for according to vehicle image AIMinimum hash mfj(AI) obtain inquiry picture A1It is minimum to breathe out
Uncommon value and warehouse-in picture A2The position of collision of minimum hash；
Offset space sets up unit, and for setting up an offset space, the offset space divides the space into M × M region,
M is the regulation parameter of setting；
Map unit, for picture A will be inquired about1Minimum hash and warehouse-in picture A2The position of collision of minimum hash is mapped to
In offset space, and statistical query picture A1Minimum hash and warehouse-in picture A2Minimum hash each area in the offset space
Collision frequency in domain；
Marking unit, is inquiry picture A for the collision frequency according to statistics1With warehouse-in picture A2Similarity given a mark；
Matching unit, for drawing inquiry picture A according to the result of marking1With warehouse-in picture A2Rapid matching retrieval result.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610916925.XA CN106528662A (en) | 2016-10-20 | 2016-10-20 | Quick retrieval method and system of vehicle image on the basis of feature geometric constraint |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610916925.XA CN106528662A (en) | 2016-10-20 | 2016-10-20 | Quick retrieval method and system of vehicle image on the basis of feature geometric constraint |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN106528662A true CN106528662A (en) | 2017-03-22 |
Family
ID=58332941
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610916925.XA Pending CN106528662A (en) | 2016-10-20 | 2016-10-20 | Quick retrieval method and system of vehicle image on the basis of feature geometric constraint |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106528662A (en) |
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107423379A (en) * | 2017-07-13 | 2017-12-01 | 西安电子科技大学 | Image search method based on CNN feature words trees |
| CN107481232A (en) * | 2017-08-17 | 2017-12-15 | 上海华力微电子有限公司 | Graphic Pattern Matching method |
| 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 |
| CN107885764A (en) * | 2017-09-21 | 2018-04-06 | 银江股份有限公司 | Based on the quick Hash vehicle retrieval method of multitask deep learning |
| CN108197538A (en) * | 2017-12-21 | 2018-06-22 | 浙江银江研究院有限公司 | A kind of bayonet vehicle searching system and method based on local feature and deep learning |
| CN108536769A (en) * | 2018-03-22 | 2018-09-14 | 深圳市安软慧视科技有限公司 | Image analysis method, searching method and device, computer installation and storage medium |
| CN109583408A (en) * | 2018-12-07 | 2019-04-05 | 高新兴科技集团股份有限公司 | A kind of vehicle key point alignment schemes based on deep learning |
| CN110147876A (en) * | 2019-05-09 | 2019-08-20 | 中山大学 | The neural network and its movement motion generation method of view-based access control model characteristic similarity |
| CN110490789A (en) * | 2019-07-15 | 2019-11-22 | 上海电力学院 | A kind of image hashing acquisition methods based on color and structure feature |
| CN116414867A (en) * | 2023-06-12 | 2023-07-11 | 中南大学 | Space-time data retrieval method based on quantization hash coding |
| CN116861023A (en) * | 2023-09-01 | 2023-10-10 | 北京达美盛软件股份有限公司 | Method and system for determining same geometric objects in three-dimensional space |
| CN117094343A (en) * | 2023-10-19 | 2023-11-21 | 成都新西旺自动化科技有限公司 | QR code decoding system and method |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102542058A (en) * | 2011-12-29 | 2012-07-04 | 天津大学 | Hierarchical landmark identification method integrating global visual characteristics and local visual characteristics |
| CN103324935A (en) * | 2013-06-27 | 2013-09-25 | 中山大学 | Method for vehicle positioning and region segmenting in image |
| CN105550657A (en) * | 2015-12-23 | 2016-05-04 | 北京化工大学 | Key point based improved SIFT human face feature extraction method |
| CN105608230A (en) * | 2016-02-03 | 2016-05-25 | 南京云创大数据科技股份有限公司 | Image retrieval based business information recommendation system and image retrieval based business information recommendation method |
| CN105975941A (en) * | 2016-05-31 | 2016-09-28 | 电子科技大学 | Multidirectional vehicle model detection recognition system based on deep learning |
| CN106033426A (en) * | 2015-03-11 | 2016-10-19 | 中国科学院西安光学精密机械研究所 | A latent semantic min-Hash-based image retrieval method |
-
2016
- 2016-10-20 CN CN201610916925.XA patent/CN106528662A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102542058A (en) * | 2011-12-29 | 2012-07-04 | 天津大学 | Hierarchical landmark identification method integrating global visual characteristics and local visual characteristics |
| CN103324935A (en) * | 2013-06-27 | 2013-09-25 | 中山大学 | Method for vehicle positioning and region segmenting in image |
| CN106033426A (en) * | 2015-03-11 | 2016-10-19 | 中国科学院西安光学精密机械研究所 | A latent semantic min-Hash-based image retrieval method |
| CN105550657A (en) * | 2015-12-23 | 2016-05-04 | 北京化工大学 | Key point based improved SIFT human face feature extraction method |
| CN105608230A (en) * | 2016-02-03 | 2016-05-25 | 南京云创大数据科技股份有限公司 | Image retrieval based business information recommendation system and image retrieval based business information recommendation method |
| CN105975941A (en) * | 2016-05-31 | 2016-09-28 | 电子科技大学 | Multidirectional vehicle model detection recognition system based on deep learning |
Non-Patent Citations (2)
| Title |
|---|
| 李珂辉: "基于非参数聚类的最小哈希函数簇构造方法", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
| 罗斌 等: "复杂环境下基于角点回归的全卷积神经网络的车牌定位", 《数据采集与处理 JOURNAL OF DATA ACQUISITION AND PROCESSING》 * |
Cited By (20)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107423379B (en) * | 2017-07-13 | 2019-10-11 | 西安电子科技大学 | Image search method based on CNN feature words tree |
| CN107423379A (en) * | 2017-07-13 | 2017-12-01 | 西安电子科技大学 | Image search method based on CNN feature words trees |
| CN107481232A (en) * | 2017-08-17 | 2017-12-15 | 上海华力微电子有限公司 | Graphic Pattern Matching method |
| CN107885764A (en) * | 2017-09-21 | 2018-04-06 | 银江股份有限公司 | Based on the quick Hash vehicle retrieval method of multitask deep learning |
| 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 |
| CN108197538A (en) * | 2017-12-21 | 2018-06-22 | 浙江银江研究院有限公司 | A kind of bayonet vehicle searching system and method based on local feature and deep learning |
| CN108197538B (en) * | 2017-12-21 | 2020-10-16 | 浙江银江研究院有限公司 | Bayonet vehicle retrieval system and method based on local features and deep learning |
| CN108536769A (en) * | 2018-03-22 | 2018-09-14 | 深圳市安软慧视科技有限公司 | Image analysis method, searching method and device, computer installation and storage medium |
| CN109583408A (en) * | 2018-12-07 | 2019-04-05 | 高新兴科技集团股份有限公司 | A kind of vehicle key point alignment schemes based on deep learning |
| CN110147876A (en) * | 2019-05-09 | 2019-08-20 | 中山大学 | The neural network and its movement motion generation method of view-based access control model characteristic similarity |
| CN110147876B (en) * | 2019-05-09 | 2023-01-03 | 中山大学 | Neural network based on visual feature similarity and action proposal generation method thereof |
| CN110490789A (en) * | 2019-07-15 | 2019-11-22 | 上海电力学院 | A kind of image hashing acquisition methods based on color and structure feature |
| CN110490789B (en) * | 2019-07-15 | 2023-03-28 | 上海电力学院 | Image abstract obtaining method based on color and structural features |
| CN116414867A (en) * | 2023-06-12 | 2023-07-11 | 中南大学 | Space-time data retrieval method based on quantization hash coding |
| CN116414867B (en) * | 2023-06-12 | 2023-08-22 | 中南大学 | Space-time data retrieval method based on quantization hash coding |
| CN116861023A (en) * | 2023-09-01 | 2023-10-10 | 北京达美盛软件股份有限公司 | Method and system for determining same geometric objects in three-dimensional space |
| CN116861023B (en) * | 2023-09-01 | 2023-12-12 | 北京达美盛软件股份有限公司 | Method and system for determining same geometric objects in three-dimensional space |
| CN117094343A (en) * | 2023-10-19 | 2023-11-21 | 成都新西旺自动化科技有限公司 | QR code decoding system and method |
| CN117094343B (en) * | 2023-10-19 | 2023-12-29 | 成都新西旺自动化科技有限公司 | QR code decoding system and method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106528662A (en) | Quick retrieval method and system of vehicle image on the basis of feature geometric constraint | |
| Hausler et al. | Patch-netvlad: Multi-scale fusion of locally-global descriptors for place recognition | |
| CN106227851B (en) | The image search method of depth of seam division search based on depth convolutional neural networks | |
| CN107885764B (en) | Rapid Hash vehicle retrieval method based on multitask deep learning | |
| CN107577990B (en) | Large-scale face recognition method based on GPU (graphics processing Unit) accelerated retrieval | |
| Lu et al. | Object detection based on SSD-ResNet | |
| CN111553201B (en) | Traffic light detection method based on YOLOv3 optimization algorithm | |
| CN102693299B (en) | System and method for parallel video copy detection | |
| CN108197538A (en) | A kind of bayonet vehicle searching system and method based on local feature and deep learning | |
| CN107908646B (en) | Image retrieval method based on hierarchical convolutional neural network | |
| CN107122712B (en) | Palm print image identification method based on CNN and bidirectional VLAD | |
| CN104090972A (en) | Image feature extraction and similarity measurement method used for three-dimensional city model retrieval | |
| CN104966081B (en) | Spine image-recognizing method | |
| JP2016062610A (en) | Feature model creation method and feature model creation device | |
| CN111046732B (en) | Pedestrian re-recognition method based on multi-granularity semantic analysis and storage medium | |
| CN107832335B (en) | Image retrieval method based on context depth semantic information | |
| CN106557579A (en) | A kind of vehicle model searching system and method based on convolutional neural networks | |
| EP3033713A1 (en) | Recognition process of an object in a query image | |
| CN105654122B (en) | Based on the matched spatial pyramid object identification method of kernel function | |
| CN105183795B (en) | Remote Sensing Imagery Change Detection information retrieval method based on content | |
| CN106874421A (en) | Image search method based on self adaptation rectangular window | |
| CN110458175A (en) | It is a kind of based on words tree retrieval unmanned plane Image Matching to selection method and system | |
| CN113920360A (en) | Road point cloud rod extraction and multi-scale identification method | |
| Li et al. | Combining topological and view-based features for 3D model retrieval | |
| CN105183845A (en) | ERVQ image indexing and retrieval method in combination with semantic features |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | 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 |
Application publication date: 20170322 |
|
| RJ01 | Rejection of invention patent application after publication |