CN103530377B - A kind of scene information searching method based on binary features code - Google Patents

A kind of scene information searching method based on binary features code Download PDF

Info

Publication number
CN103530377B
CN103530377B CN201310483341.4A CN201310483341A CN103530377B CN 103530377 B CN103530377 B CN 103530377B CN 201310483341 A CN201310483341 A CN 201310483341A CN 103530377 B CN103530377 B CN 103530377B
Authority
CN
China
Prior art keywords
step
image
hash
information
descriptor
Prior art date
Application number
CN201310483341.4A
Other languages
Chinese (zh)
Other versions
CN103530377A (en
Inventor
桂振文
刘越
王涌天
Original Assignee
北京理工大学
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by 北京理工大学 filed Critical 北京理工大学
Priority to CN201310483341.4A priority Critical patent/CN103530377B/en
Publication of CN103530377A publication Critical patent/CN103530377A/en
Application granted granted Critical
Publication of CN103530377B publication Critical patent/CN103530377B/en

Links

Abstract

The present invention provides a kind of scene information searching method based on binary features code, belongs to mobile augmented reality technical field, and detailed process is: step 1: gathers the image to be identified of current scene, and obtains GPS information and gravity direction information;Step 2: the descriptor characteristic vector of image to be identified;Step 3: GPS information, gravity direction information and descriptor characteristic vector are packaged into a descriptor file and are sent to server;Step 4: server calculates the angle between principal direction and the gravity direction of descriptor characteristic vector;Step 5: by descriptor characteristic vector, uses Hash function to map;Step 6: search chained list GPS information corresponding to closest with GPS information;Step 7: acquire Hash table collection;Step 8: Hash table collection is filtered based on described angle;Step 9: concentrate being filtrated to get Hash table, find matched sample image, scene information corresponding for this sample image is returned to mobile terminal.

Description

A kind of scene information searching method based on binary features code

Technical field

The invention belongs to mobile augmented reality technical field, be specifically related to a kind of field based on binary features code Scape information search method.

Background technology

Picture search is the hot issue of content-based information retrieval research, at remote sensing image processing, medical science figure As the fields such as process and augmented reality have a wide range of applications.

At present, along with the development of internet, the mankind are stepping into an informationalized society, and internet is Become the Important Platform that the mankind issue, obtain, exchange information.Internet information volume index level ground increases, Make how to allow user that its information needed can be found rapidly and accurately in the data of magnanimity to become a weight The problem wanted.In the last few years, camera, smart mobile phone, PAD, the electronic product such as video camera universal, I Anytime anywhere can take the scenery that we like, animal, food, etc. various pictures. By the end of in January, 2010, Facebook claims that the picture number on its website alreadys more than 25,000,000,000.Face Picture resource to such magnanimity, finding our picture interested the most fast and accurately is that we have to The problem solved, is an important research direction of business circles and academia.But, along with picture scale Greatly increase, it is ensured that the real-time of picture search, corresponding Image Coding, image retrieval technologies and data Storehouse index technology also must do corresponding adjustment or acceleration.

The fast development of simultaneous computer soft and hardware technology, for augmented reality walk out indoor application and then Support that complicated analysis, decision-making and management lays a solid foundation.Some mobile terminal devices (as PDA, Smart mobile phone etc.) function more and more abundanter, and have embedded OS, touch-screen, GPS Location, the first-class function of video camera, also possessed stronger calculating and disposal ability simultaneously.These functions It is integrated into exploitation augmented reality system based on mobile terminal to lay a good foundation.According to relevant information, by the end of 2010 Year, China mobile phone user was up to 7.4 hundred million, and the user wherein having smart mobile phone account for suitable proportion, Smart mobile phone will have the biggest application potential as the application platform of augmented reality.3G net progressively open, Run, it is meant that the beginning of mobile value-added service new era, augmented reality and LBS combine permissible Realize the real-time, interactive of information, dynamic three-dimensional display, man-machine interface can be made more friendly and have intelligent.

Based on above-mentioned analysis, in conjunction with having camera, GPS sensor and the terminal of gravity sensor and service The image recognition of device end and matching technique, can the ONLINE RECOGNITION of the extensive object of scene etc be become can Energy.

The research work of the many image recognitions before but is all in the case of sample size is not very big Consider that image retrieval problem, many methods can not be generalized to more massive problem of image recognition, systematicness There is limitation in the data scale that can and can process.When the data of City-level scale, during with million for unit, Image identification system needs the memory space of magnanimity and the quick computing capability of mass data.Because image itself Need the biggest space to store, be also required to the storage of substantial amounts of space from the various feature interpretation vectors of image zooming-out. Meanwhile, in image recognition and matching process, descriptor index, coupling are also required to powerful calculating ability.

Summary of the invention

In view of this, the invention provides a kind of scene information searching method based on binary features code, should Method is capable of the identification of online large nuber of images, decreases memory data output simultaneously, improves image recognition Rate and the speed of retrieval.

Realize technical scheme as follows:

A kind of scene information searching method based on binary features code, the sample image storehouse being suitable for the method is full Three conditions of foot: the most each sample image is with GPS information, and the most each sample image uses binary features Representation, and store into the chained list with GPS information as cluster centre, 3. sample image storehouse correspondence one determine, For mapping the Hash function of binary features code;The detailed process of the method is:

Step 1: utilize the capture apparatus of terminal to gather the image to be identified of current scene, and obtain current scene GPS information and terminal gather image time gravity direction information;

Step 2: use local feature detection algorithm that image to be identified is carried out feature point detection, utilize feature to retouch State symbol detected characteristic point is described, it is thus achieved that descriptor characteristic vector;

Step 3: described GPS information, gravity direction information and descriptor characteristic vector are packaged into one and retouch State symbol file cocurrent and give server;

Step 4: server extracts gravity direction information and descriptor characteristic vector from the descriptor file received, And calculate the angle between principal direction and the gravity direction of descriptor characteristic vector;

Step 5: by the descriptor characteristic vector of image to be identified, use the Hash corresponding with sample image storehouse Function carries out Hash mapping, obtains Hash bit string to be matched, then Hash bit string to be matched is divided into 8 groups;

Step 6: according to described GPS information, searches closest with this GPS information from sample image storehouse The chained list corresponding to GPS information;

Step 7: for each group in the 8 groups of Hash bit strings obtained in step 5, obtain in step 6 Chained list is searched and differs from it by the Hash table less than or equal to Hamming distance threshold value, and to the Hash found Table carries out union and obtains Hash table collection;

Step 8: to and operate the Hash table collection that obtains, i.e. binary code collection, the folder obtained according to step 4 It is filtered by angle;

Step 9: concentrate being filtrated to get Hash table, statistics is with image binary code number of matches to be identified Many sample images, then return to mobile terminal by scene information corresponding for this sample image.

Further, the present invention, before detecting image characteristic point to be identified, also includes dropping image Sampling processing.

Further, the nearest GPS information that the present invention finds when step 6 is corresponding with image to be identified When GPS information apart exceedes setting threshold value, now server generates the feedback that cannot inquire associated scenario information Signal return is to mobile terminal.

Further, the sample image storehouse that the present invention is suitable for uses following methods to set up:

S01, obtains the sample image of band GPS information;

S02, extracts the local feature of every width sample image, and described local feature is converted to descriptor feature Vector, utilizes described descriptor characteristic vector to train Hash function, determines the parameter of Hash function, then Descriptor maps feature vectors is Hash bit string by utilization training obtains, the Hash function of parameter determination, This Hash bit string is referred to as binary features code;

S03, sets up the index of higher with the GPS of sample image as cluster centre, by affiliated same cluster The binary features code that the sample image at center is corresponding stores in a chained list.

Beneficial effect:

The first, the present invention uses the camera of mobile terminal that current scene carries out IMAQ, and utilize right Characteristics of image binary coding compress technique (i.e. Hash mapping) and computer vision technique (i.e. Hamming distance Identify), collected outdoor scene is identified, relies on internet or other means of communication, it is provided that be identified The various information of scene, it is also possible to additional further various application in these information, thus be user The relevant information obtaining this scene provides a kind of more convenient means.

The second, the hash of present invention Hash hash function represents, it is only necessary to tens just represent retouching of higher-dimension State vector, save memory space, provide feasibility for large-scale storage data.Use binary segmentation Searching algorithm improves the Rapid matching ability of magnanimity descriptor, utilizes the two main sides entering feature descriptor simultaneously Carry out further non-similarity filtration to the angle of gravity direction, make binary system descriptor similarity search more Accurately, the robust control policy for outdoor massive picture provides condition advantageously.

3rd, the present invention, when carrying out similar image lookup for current task, first judges according to GPS information Whether there is the sample image that position is nearer, if it has, carry out follow-up binary features code similarity system design, If it is not, directly notice mobile terminal can not find similar sample image;Therefore the present invention passes through GPS The positional information of sensor, reduces the scope of matched sample, saves the time of coupling.

4th, GPS value, when recording sample binary features code, is clustered, chooses cluster by the present invention The position of this class of center representative, and by this kind of sampling feature vectors, be placed in a big chained list, convenient Make a look up.

5th, the present invention can extend the interactive application of intelligent terminal, meet tourism, navigation, traffic, The expanded application on intelligent terminal such as hotel service, enables Virtual network operator and content supplier to utilize it Abundant server resource and superior server performance develop its business.

Accompanying drawing explanation

Fig. 1 is the scene information search principle figure of binary features code;

Fig. 2 is many index structures figure of the outdoor scene training sample vector of binary features of the present invention;

Fig. 3 is gravity direction and feature principal direction angle;

Fig. 4 is the code pattern of sensor information;

Fig. 5 is packet binary system similarity search schematic diagram.

Detailed description of the invention

Below in conjunction with drawings and Examples, the present invention is described in detail.

As it is shown in figure 1, present invention scene information based on binary features code searching method, it is suitable for the method Sample image storehouse meet three conditions: the most each sample image with GPS information, the most each sample image Use binary features representation, and store into the chained list with GPS information as cluster centre, 3. sample image That storehouse correspondence one determines, for mapping the Hash function of binary features code;The detailed process of the method is: Step 1: the capture apparatus that user opens a terminal, terminal gathers the image to be identified of current scene;Recall end The GPS sensor interface of end and gravity sensor interface, obtain GPS information and the terminal collection of current scene Gravity direction information g=[g during imagex,gy,gz], wherein gx,gy,gzObtain for terminal acceleration of gravity inductor The acceleration of three change in coordinate axis direction obtained.

Step 2: in order to reduce operand, terminal carries out down-sampled process to the image to be identified collected, with Reduce image resolution ratio;Then local feature detection algorithm (can be such as SIFT, SURF or ORB) is used Image to be identified carries out feature point detection, and recycling feature descriptor (can be such as FREAK, Fast Retina Keypoint) detected characteristic point is described, it is thus achieved that descriptor characteristic vector, thus realize The local feature of image to be identified is converted into descriptor characteristic vector.It is big owing to piece image may comprise The information of amount, therefore piece image may have up to a hundred descriptor characteristic vectors.

This step detects characteristic point and acquisition descriptor characteristic vector realizes for employing prior art, therefore exist This is not described in detail.

Step 3: described GPS information, gravity direction information and descriptor characteristic vector are packaged into one and retouch State symbol file cocurrent and give server.

The position of image to be identified is obtained, by GPS information in descriptor file for the ease of far-end server Being placed on beginning, this descriptor file can be sent to far-end server by wireless network.

Step 4: server, from the descriptor file received, extracts gravity direction information and descriptor feature Vector, and calculate the angle between principal direction and the gravity direction of descriptor characteristic vector.

As it is shown on figure 3, because outdoor shooting building scenes exists the object of a large amount of analog structure, depend merely on office Portion's feature cannot be distinguished by the analog structure inside image.But, by terminal appearance when representing shooting image to be identified Angle between gravity direction and the principal direction of descriptor characteristic vector of gesture can preferably filter similar features; Therefore obtaining angle between the two in this step is that subsequent step 8 is prepared.Meanwhile, as shown in Figure 4 Coded system, encodes angle, such as, be 11111 when angle is 30 degree;It is 11110 when 60 degree; When being 90 degree 11100;When being 120 degree 11000;When 150 degree 10000;It is 00000 when 180 degree, from And realize being expressed as angle the form of binary code.

For each terminal device with gravity sensing, the computational methods of the gravity direction of its correspondence are all true Fixed, as a example by certain Mobile phone, the computational methods of gravity direction are illustrated below:

Such as: image midpoint p=[u, v, 1]TThe gravity direction vector at place is d=p '-p;

Wherein d=[du,dv,0]TFor the gravity direction at a p, p ' is calculated by following formula:

P '=[wu ', wv ', w]T=p+Kg

Wherein g=[gx,gy,gz]TIt is the acceleration of three change in coordinate axis direction that acceleration of gravity inductor obtains in mobile phone Degree vector, K is the internal reference matrix of mobile phone camera.

Can be calculated the angle of gravity direction in image by d is θg=arctan (dv/du)。

Step 5: by the descriptor characteristic vector of image to be identified, use the Hash corresponding with sample image storehouse Function carries out Hash mapping, obtains Hash bit string to be matched, and this Hash bit string is referred to as binary features code, Again Hash bit string to be matched is divided into 8 groups;

Step 6: according to described GPS information, searches closest with this GPS information from sample image storehouse GPS information corresponding to chained list;

Step 7: for each group in the 8 groups of Hash bit strings obtained in step 5, obtain in step 6 Chained list is searched and differs from it by the Hash table less than or equal to Hamming distance threshold value, and to the Hash found Table carries out union and obtains Hash table collection.The above-mentioned 8 groups of bit strings obtaining packet make a look up as carrying out parallel 's.

Generally Hamming distance threshold value is: if whole Hash bit string code length to be matched is 64, Hamming distance Threshold value is 8, then the bit string Hamming distance threshold value of packet is 1, if whole Hash bit string code to be matched A length of 128, Hamming distance threshold value is 16, then the threshold value that block code compares is 2.

Hamming distance be calculated as prior art, an example is set forth below it is briefly described:

As shown in Figure 5, it is assumed that Hash bit string length L=128 position to be matched, the Hamming distance threshold between bit string Value is set to 32, when Hamming distance is less than or equal to 32, then it is assumed that two bit strings are similar, i.e. L1XOR L2 ≤ 32, dissimilar more than then thinking, i.e. L1XOR L2 > 32.Again this bit string is grouped, it is assumed that be divided into 8 Group sub1, sub2 ..., sub8, often group 16, then the Hamming distance of two bit strings is calculated as following formula:

Han min gdis tan ce = Σ 1 8 | | su b i - sub i ′ | | ≤ 32

||subi-subi' | | for the Hamming distance computing formula of subgroup, then at least there is one group of Chinese in 8 packets Prescribed distance is less than 32/8=4 position, is following formula:

||subi-subi′||i≤4,i=1,…,8

The present invention method of Hash bit string (i.e. binary features code) packet index, enters according to the two of packet At least there is a figure place obtained less than original code length/packet count in condition code processed, sets Querying by group Hamming Distance is the value of code length/packet count, is first grouped sub-bit string to 8, searches all of sub-bit string and indexes at Hamming Binary features code chained list within distance threshold, then binary features code mark in taking-up chained list is carried out union (∪), in the middle of removal, there is duplicate marking in computing;To ∪ gather, filter with angle, remaining all It it is the similar feature of gravity direction principal direction angle;Finally, by the mark of binary features code, take out two 128 all-keys of system feature and the all-key of query characteristics, carry out Hamming one by one and compare, and distance is the shortest to be Similar, this is the principle of the present invention.

Step 8: to and operate the Hash table collection that obtains, i.e. binary code collection, the folder obtained according to step 4 It is filtered by angle, will concentrate expression sample image gravity direction and descriptor characteristic vector by binary code The binary code of the angle of principal direction binary code with image to be identified one by one carries out Hamming distance calculating, right In the Hamming distance binary code more than or equal to 2, it is all that angle difference is bigger, all filters tune.

Step 9: concentrate at the binary code being filtrated to get, statistics and image binary code number of matches to be identified Most sample images, and as result images, it is then back to the scene information (example that result images is corresponding Such as market situation that this information is scene periphery, traffic conditions etc.) to mobile terminal;Terminal can show Stating recognition result, user can click on above-mentioned classification results, checks details.

So far, this flow process terminates.

The sample image storehouse that scene information searching method of the present invention is suitable for can use following steps to set up:

S01, obtains the sample image of band GPS information;

In general can by scene obtain image, such as from network download or shoot on the spot, each scene from Different angles obtain a few width sample images, and corresponding to sample image, the GPS information of scene is exactly sample image GPS information.

S02, extracts the local feature of every width sample image, and described local feature is converted to descriptor feature Vector, utilizes described descriptor characteristic vector to train Hash function, determines the parameter of Hash function, then Descriptor maps feature vectors is Hash bit string by the Hash function of the parameter determination that training obtains, this Hash Bit string is referred to as binary features code.

Hash function described in this step preferably selects Binary LSH(Locality Sensitive Hashing).

S03, sets up the index of higher with the GPS of sample image as cluster centre, by affiliated same cluster The binary features code (Hash bit string) that the sample image at center is corresponding stores in a chained list, such as Fig. 2 Shown in, in the chained list of Fig. 2, the corresponding sample image of each Index note, its first row form stores Sample image binary features code information, secondary series is used for storing ID and the GPS information of sample image, the Three row may be used for storing relevant scene information of image etc..

In sum, these are only presently preferred embodiments of the present invention, be not intended to limit the guarantor of the present invention Protect scope.All within the spirit and principles in the present invention, any modification, equivalent substitution and improvement etc. made, Should be included within the scope of the present invention.

Claims (4)

1. a scene information searching method based on binary features code, is suitable for the sample image storehouse of the method Meet three conditions: the most each sample image is with GPS information, and the most each sample image uses binary system special Levying representation, and store into the chained list with GPS information as cluster centre, 3. sample image storehouse correspondence one determines , for mapping the Hash function of binary features code;It is characterized in that, the detailed process of the method is:
Step 1: utilize the capture apparatus of terminal to gather the image to be identified of current scene, and obtain current scene GPS information and terminal gather image time gravity direction information;
Step 2: use local feature detection algorithm that image to be identified is carried out feature point detection, utilize feature to retouch State symbol detected characteristic point is described, it is thus achieved that descriptor characteristic vector;
Step 3: described GPS information, gravity direction information and descriptor characteristic vector are packaged into one Descriptor file also sends;
Step 4: server extract from the descriptor file received gravity direction information and descriptor feature to Amount, and calculate the angle between principal direction and the gravity direction of descriptor characteristic vector;
Step 5: by the descriptor characteristic vector of image to be identified, use the Hash corresponding with sample image storehouse Function carries out Hash mapping, obtains Hash bit string to be matched, then Hash bit string to be matched is divided into 8 groups;
Step 6: according to described GPS information, searches closest with this GPS information from sample image storehouse The chained list corresponding to GPS information;
Step 7: for each group in the 8 groups of Hash bit strings obtained in step 5, obtain in step 6 Chained list is searched and differs from it by the Hash table less than or equal to Hamming distance threshold value, and to the Hash found Table carries out union and obtains Hash table collection;
Step 8: to and operate the Hash table collection that obtains, i.e. binary features code collection, obtains according to step 4 Angle it is filtered;
Step 9: concentrating being filtrated to get Hash table, statistics mates number with image binary features code to be identified Measure most sample images, then scene information corresponding for this sample image is returned to mobile terminal.
The most according to claim 1, scene information searching method based on binary features code, its feature exists In, before image characteristic point to be identified is detected, also include image is carried out down-sampled process.
The most according to claim 1, scene information searching method based on binary features code, its feature exists The GPS information corresponding with image to be identified in, the nearest GPS information that finds when step 6 apart exceedes When setting threshold value, now server generation cannot inquire the feedback signal of associated scenario information and return to mobile Terminal.
4. according to scene information searching method based on binary features code described in claim 1,2 or 3, its Being characterised by, described sample image storehouse uses following methods to set up:
S01, obtains the sample image of band GPS information;
S02, extracts the local feature of every width sample image, and described local feature is converted to descriptor feature Vector, utilizes described descriptor characteristic vector to train Hash function, determines the parameter of Hash function, then Descriptor maps feature vectors is Hash bit string by utilization training obtains, the Hash function of parameter determination, This Hash bit string is referred to as binary features code;
S03, sets up the index of higher with the GPS of sample image as cluster centre, by affiliated same cluster The binary features code that the sample image at center is corresponding stores in a chained list.
CN201310483341.4A 2013-10-16 2013-10-16 A kind of scene information searching method based on binary features code CN103530377B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201310483341.4A CN103530377B (en) 2013-10-16 2013-10-16 A kind of scene information searching method based on binary features code

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201310483341.4A CN103530377B (en) 2013-10-16 2013-10-16 A kind of scene information searching method based on binary features code

Publications (2)

Publication Number Publication Date
CN103530377A CN103530377A (en) 2014-01-22
CN103530377B true CN103530377B (en) 2016-08-24

Family

ID=49932386

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201310483341.4A CN103530377B (en) 2013-10-16 2013-10-16 A kind of scene information searching method based on binary features code

Country Status (1)

Country Link
CN (1) CN103530377B (en)

Families Citing this family (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104572986A (en) * 2015-01-04 2015-04-29 百度在线网络技术(北京)有限公司 Information searching method and device
CN104616017A (en) * 2015-02-04 2015-05-13 四川中科腾信科技有限公司 Intelligent image recognition method
CN105022808A (en) * 2015-06-29 2015-11-04 程文举 Binary constant value interval matching method
CN106446948B (en) * 2016-09-23 2019-04-16 西安交通大学 A kind of mobile learning scene perception method based on sensing data Yu learning manipulation behavior

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP2297703A1 (en) * 2008-06-03 2011-03-23 ETH Zurich Method and system for generating a pictorial reference database using geographical information
CN102663448A (en) * 2012-03-07 2012-09-12 北京理工大学 Network based augmented reality object identification analysis method
CN102819752A (en) * 2012-08-16 2012-12-12 北京理工大学 System and method for outdoor large-scale object recognition based on distributed inverted files
CN102831405A (en) * 2012-08-16 2012-12-19 北京理工大学 Method and system for outdoor large-scale object identification on basis of distributed and brute-force matching
EP2630575A1 (en) * 2010-10-22 2013-08-28 Telefonaktiebolaget L M Ericsson (publ) Image matching apparatus and image matching method

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP2297703A1 (en) * 2008-06-03 2011-03-23 ETH Zurich Method and system for generating a pictorial reference database using geographical information
EP2630575A1 (en) * 2010-10-22 2013-08-28 Telefonaktiebolaget L M Ericsson (publ) Image matching apparatus and image matching method
CN102663448A (en) * 2012-03-07 2012-09-12 北京理工大学 Network based augmented reality object identification analysis method
CN102819752A (en) * 2012-08-16 2012-12-12 北京理工大学 System and method for outdoor large-scale object recognition based on distributed inverted files
CN102831405A (en) * 2012-08-16 2012-12-19 北京理工大学 Method and system for outdoor large-scale object identification on basis of distributed and brute-force matching

Non-Patent Citations (4)

* Cited by examiner, † Cited by third party
Title
City-Scale Landmark Identification on Mobile Devices;David M.Chen;《2011 IEEE conference on computer vision and pattern recognition》;20111203;全文 *
Outdoor Landmark-view Recognition Based on Bipartite-graph;Eduardo Todt;《proceedings of the 2007 ieee international conference on robotics and automation》;20070414;全文 *
基于移动平台增强现实系统的研究;孔祥;《中国优秀硕士学位论文全文数据库》;20130430;全文 *
移动增强现实中的快速场景识别与注册方法研究;毛谭;《中国优秀硕士学位论文全文数据库》;20130731;全文 *

Also Published As

Publication number Publication date
CN103530377A (en) 2014-01-22

Similar Documents

Publication Publication Date Title
Chen et al. Event detection from flickr data through wavelet-based spatial analysis
Li et al. Location recognition using prioritized feature matching
JP5654127B2 (en) Object recognition using incremental feature extraction
US9600499B2 (en) System for collecting interest graph by relevance search incorporating image recognition system
US8768107B2 (en) Matching an approximately located query image against a reference image set
Li et al. Worldwide pose estimation using 3d point clouds
US20130243250A1 (en) Location of image capture device and object features in a captured image
EP2054855B1 (en) Automatic classification of objects within images
Zheng et al. Packing and padding: Coupled multi-index for accurate image retrieval
RU2533441C2 (en) Method and apparatus for facilitating content-based image search
KR101895647B1 (en) Location-aided recognition
Zhang et al. A fine-grained image categorization system by cellet-encoded spatial pyramid modeling
US8775401B2 (en) Shape based picture search
US9036925B2 (en) Robust feature matching for visual search
Luo et al. Geotagging in multimedia and computer vision—a survey
US9230172B2 (en) Systems and methods for image-feature-based recognition
US10409855B2 (en) System and method of identifying visual objects
Vo et al. A survey of fingerprint-based outdoor localization
Baatz et al. Leveraging 3d city models for rotation invariant place-of-interest recognition
US7860317B2 (en) Generating search results based on duplicate image detection
Wang et al. Effective multi-query expansions: Collaborative deep networks for robust landmark retrieval
CN105900084A (en) Systems, methods, and apparatus for image retrieval
US20110184953A1 (en) On-location recommendation for photo composition
JP5461523B2 (en) Image Classification using shooting position sequence information
JP2007206920A (en) Image processor and image processing method, retrieving device and method, program and recording medium

Legal Events

Date Code Title Description
PB01 Publication
C06 Publication
SE01 Entry into force of request for substantive examination
C10 Entry into substantive examination
GR01 Patent grant
C14 Grant of patent or utility model