CN107220325A - A kind of similar icon search methods of APP based on convolutional neural networks and system - Google Patents
A kind of similar icon search methods of APP based on convolutional neural networks and system Download PDFInfo
- Publication number
- CN107220325A CN107220325A CN201710366244.5A CN201710366244A CN107220325A CN 107220325 A CN107220325 A CN 107220325A CN 201710366244 A CN201710366244 A CN 201710366244A CN 107220325 A CN107220325 A CN 107220325A
- Authority
- CN
- China
- Prior art keywords
- app
- sample
- target
- convolutional neural
- neural networks
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
- G06F16/53—Querying
- G06F16/532—Query formulation, e.g. graphical querying
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
- G06F16/58—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
- G06F16/583—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content
- G06F16/5838—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content using colour
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Library & Information Science (AREA)
- Databases & Information Systems (AREA)
- Mathematical Physics (AREA)
- Biomedical Technology (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Evolutionary Computation (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of similar icon search methods of APP based on convolutional neural networks and system, the realization of wherein method includes：By sampling feature vectors, sample file mark and N number of index deposit searching system；For target APP icons, target feature vector is divided into N number of target part, each target part is combined, M objective cross characteristic vector is obtained；To M assemblage characteristic vector, M retrieval is carried out in searching system, M file identification set is obtained, by M file identification collection conjunction union, and concentrate and linearly calculated, obtain the similarity of sample APP icons and target APP icons, sample APP icons be ranked up using similarity.Accuracy rate is low in being retrieved instant invention overcomes traditional images, and the problem of recall precision is not high, while providing the user very big convenience, can search for corresponding application program of mobile phone according to icon.
Description
Technical field
The invention belongs to field of data retrieval, more particularly, to a kind of similar icons of APP based on convolutional neural networks
Search method and system.
Background technology
With the development and the raising of smart mobile phone popularity rate of internet, hand-held mobile terminal has obtained huge hair
Exhibition, incident is the increase of mobile phone terminal number of applications, and hand is passed through for user and safety detection personnel
Machine application icon is come quickly to search out corresponding or similar application program of mobile phone be an extremely difficult thing.
In this regard, needing to use image retrieval technologies.
Current image retrieval technologies are concentrated mainly on text based retrieval and content-based retrieval.Based on text
Retrieval refer to that keyword or text header and some additional informations are set up to image file, and image is described, then will
The store path of image and the keyword of image set up contact.Its shortcoming is：As the appearance of great amount of images is, it is necessary to a large amount of
Labour go to manage and annotate these images：Understanding of the different people to same piece image is different, and text description information is relative
It is subjective；The people of country variant can cause the difference of the semantic understanding to same image due to the difference of language.Thus it is based only on
The retrieval of keyword can not meet the retrieval requirement of user.Moreover, the group of traditional database retrieval result and information
The display mode for knitting mode and Query Result is relevant, and can not be exported according to the similarity degree of Query Result.Can be to many
Media data content carries out the development trend that automatic semantic analysis, expression and retrieval are database and other information systemses.
Main based on content is the feature for extracting image, and the meter of similarity is then carried out further according to characteristics of image
Calculate, algorithm more in large-scale data has D-hash, SIFT, LBP scheduling algorithm at present, but the shortcoming of these algorithms is also
Obvious, for D-hash algorithms, recall precision is higher, but accuracy rate is very low, and the algorithm is to profile
Susceptibility is too high, it is impossible to retrieved in the consistent image of profile；, should for D-hash algorithms for SIFI algorithms
Algorithm has very high accuracy rate, but the complexity of the algorithm is very high, not only when calculating SIFT feature, most importantly exists
During retrieval, SIFT feature is made up of the N number of 128 floating type vectors tieed up, and when two images carry out Similarity Measure, amount of calculation is non-
Chang great, and SIFT algorithms in large-scale data when being retrieved, and can only be examined using the method linearly calculated in full storehouse
The calculating with each image similarity of progress in full storehouse is needed during rope, which greatly enhances the time of retrieval.
As can be seen here, prior art has that recall precision is low and the low technical problem of retrieval rate.
The content of the invention
For the disadvantages described above or Improvement requirement of prior art, the invention provides a kind of based on convolutional neural networks
APP (Armor Piercing Proof, computer applied algorithm) similar icon search method and system, its object is to by sample
Eigen vector, sample file mark and index deposit searching system, to assemblage characteristic vector, are examined in searching system
Rope, obtains file identification set, by file identification collection conjunction union, and concentrate and linearly calculated, obtain sample APP icons
With the similarity of target APP icons, sample APP icons are ranked up using similarity.Thus solve prior art and there is inspection
Rope efficiency is low and the low technical problem of retrieval rate.
To achieve the above object, according to one aspect of the present invention, there is provided a kind of APP phases based on convolutional neural networks
Like logo retrieval method, including：
(1) for sample APP icons, sampling feature vectors are extracted based on convolutional neural networks, sampling feature vectors are recorded
Sample file mark, sampling feature vectors are divided into N number of sample portion, an index is set up in each sample portion, obtains
To N number of index, by sampling feature vectors, sample file mark and N number of index deposit searching system；
(2) for target APP icons, target feature vector is extracted based on convolutional neural networks, target feature vector is equal
It is divided into N number of target part, each target part is combined, obtains M objective cross characteristic vector；
(3) to M assemblage characteristic vector, M retrieval is carried out in searching system, M file identification set is obtained, by M
Individual file identification collection conjunction union, and concentrate and linearly calculated, obtain sample APP icons similar to target APP icons
Degree, is ranked up using similarity to sample APP icons.
Further, step (1) also includes pre-processing sample APP icons, and step (2) also includes to target APP
Icon is pre-processed.
Further, pretreatment includes：Image gray processing, image zooming and image normalization.
Further, step (1) also includes carrying out sampling feature vectors hashed processing, and step (2) also includes to mesh
Mark characteristic vector and carry out hashed processing.
Further, the specific implementation of hashed processing is：A threshold value is set, characteristic vector x is mapped as two
System vector f (x),The characteristic vector is any in sampling feature vectors and target feature vector
One.
It is another aspect of this invention to provide that there is provided a kind of similar icon searching systems of APP based on convolutional neural networks,
Including：
Searching system module, for for sample APP icons, sampling feature vectors, note to be extracted based on convolutional neural networks
The sample file mark of sampling feature vectors is recorded, sampling feature vectors are divided into N number of sample portion, built in each sample portion
A vertical index, obtains N number of index, by sampling feature vectors, sample file mark and N number of index deposit searching system；
Assemblage characteristic vector module, for for target APP icons, based on convolutional neural networks extract target signature to
Amount, is divided into N number of target part by target feature vector, each target part is combined, obtain M objective cross feature
Vector；
Icon order module, for M assemblage characteristic vector, M retrieval to be carried out in searching system, obtains M text
Part logo collection, by M file identification collection conjunction union, and concentrate and linearly calculated, obtain sample APP icons and target
The similarity of APP icons, is ranked up using similarity to sample APP icons.
Further, searching system module also includes pre-processing sample APP icons, and assemblage characteristic vector module is also
Including being pre-processed to target APP icons.
Further, pretreatment includes：Image gray processing, image zooming and image normalization.
Further, searching system module also includes carrying out sampling feature vectors hashed processing, assemblage characteristic vector
Module also includes carrying out hashed processing to target feature vector.
Further, the specific implementation of hashed processing is：A threshold value is set, characteristic vector x is mapped as two
System vector f (x),The characteristic vector is any in sampling feature vectors and target feature vector
One.
In general, by the contemplated above technological invention of the present invention compared with prior art, it can obtain down and show
Beneficial effect：
(1) present invention by sampling feature vectors, sample file identify and index deposit searching system, to assemblage characteristic to
Amount, is retrieved in searching system, obtains file identification set, by file identification collection conjunction union, and concentrate carry out line
Property calculate, obtain the similarity of sample APP icons and target APP icons, sample APP icons be ranked up using similarity.
And then it is low to overcome accuracy rate in traditional images retrieval, the problem of recall precision is not high, while provide the user very big convenience,
Corresponding application program of mobile phone can be searched for according to icon；Also carried simultaneously for safety detection personnel for counterfeit APP detection
A new approach is supplied；What is more important is bank or counterfeit APP or issue safety warning etc. hit in some mechanisms
There is provided technical support.
(2) preferred, the present invention is pre-processed to sample APP icons and target APP icons, by the specification of icon image
It is unitized, it is ensured that the high efficiency of calculating.
(3) preferred, the present invention carries out hashed processing to sample APP icons and target APP icons, improves retrieval effect
Rate.
(4) preferred, the present invention carries out hashed processing to sample APP icons and target APP icons, and given threshold is
0.5, while recall precision is ensured, improve retrieval rate.
Brief description of the drawings
Fig. 1 is a kind of stream of the similar icon search methods of APP based on convolutional neural networks provided in an embodiment of the present invention
Cheng Tu；
Fig. 2 is the structural representation of convolutional neural networks provided in an embodiment of the present invention；
Fig. 3 is the embodiment of the present invention by the schematic diagram that target feature vector is divided equally and changes.
Embodiment
In order that the purpose of the present invention, technological invention and advantage are more clearly understood, it is right below in conjunction with drawings and Examples
The present invention is further elaborated.It should be appreciated that the specific embodiments described herein are merely illustrative of the present invention, and
It is not used in the restriction present invention.As long as in addition, technical characteristic involved in each embodiment of invention described below
Not constituting conflict each other can just be mutually combined.
As shown in figure 1, a kind of similar icon search methods of APP based on convolutional neural networks, including：
(1) for sample APP icons, sampling feature vectors are extracted based on convolutional neural networks, sampling feature vectors are recorded
Sample file mark, sampling feature vectors are divided into N number of sample portion, an index is set up in each sample portion, obtains
To N number of index, by sampling feature vectors, sample file mark and N number of index deposit searching system；
(2) for target APP icons, target feature vector is extracted based on convolutional neural networks, target feature vector is equal
It is divided into N number of target part, each target part is combined, obtains M objective cross characteristic vector；
(3) to M assemblage characteristic vector, M retrieval is carried out in searching system, M file identification set is obtained, by M
Individual file identification collection conjunction union, and concentrate and linearly calculated, obtain sample APP icons similar to target APP icons
Degree, is ranked up using similarity to sample APP icons.
Further, sample APP icons and target APP icons are pre-processed.
When being retrieved to icon, in order to extract the characteristic vector of icon, it is necessary to by the uniform specification of image, this hair
Pretreatment in bright for icon image mainly has three steps：Image gray processing, image zooming, image normalization.
Icon image uses RGB coded system, a pixel by a three-dimensional vector representation as (R, G,
B), but be due to that the input of convolutional neural networks of design is one-dimensional, while for the high efficiency of calculating, so need will figure
The gray scale conversion algorithm that logo image used in the conversion of gray processing, the present invention is as follows：
Gray (i, j)=0.299*R (i, j)+0.578*G (i, j)+0.114*B (i, j)
(i, j) is pixel point coordinates, and Gray (i, j) is gray value when pixel point coordinates is (i, j), and R (i, j) is pixel
R passages when point coordinates is (i, j), G (i, j) is G passages when pixel point coordinates is (i, j), and B (i, j) is pixel point coordinates
Channel B during for (i, j).
For sample APP icons and target APP icons, icon image is not of uniform size, while the input of convolutional neural networks
Size is fixed, it is necessary to reduced or amplified to icon, bilinear interpolation is employed in the present invention, for a target picture
Vegetarian refreshments, obtains its floating-point coordinate for (i+u, j+v), wherein i, j is the coordinate points in original image, u, v by reciprocal transformation
For it is interval [0,1) between floating number, then the value for four points that the value of target pixel points can be closest in original image Lai
It is determined that, corresponding formula is：
Wherein, f (i, j) is the pixel value at (i, j) place in original image, the like, during f (i, j+1) is original image
The pixel value at (i, j+1) place, (i+1 is j) that (i+1, j) place pixel value, f (i+1, j+1) is original image in original image to f
In (i+1, j+1) place pixel value, four points around use calculate pixel value f (i+u, the j+ of object pixel (i+u, j+v)
v)。
3rd to do is to that image is normalized, and the present invention is using the algorithm of linear normalization, and specific formula is such as
Under：
It is 255 that min (x), which is set, as 0, max (x), then substitutes into above formula, and x is that four points around use are calculated
Go out the pixel value f (i+u, j+v) of object pixel (i+u, j+v), calculate the pixel value x after normalization.
Further, characteristic vector pickup is carried out to pretreated icon, deep learning is more traditional before occurring
Method is all the characteristic vector pickup rule manually set, present invention employs the achievement in research of forefront, uses deep learning
Technology is extracted to icon image characteristic vector, and the characteristic vector extracted is easy in large-scale data
Carry out the calculating of similarity.
Specific way is as follows：
Convolutional neural networks are designed, with reference to AlexNet network structures, addition one connects entirely between layer 7 and the 8th layer
Layer is connect, the dimension of this layer is 64, and using softmax functions as activation primitive, specific design is as shown in Figure 2.
The neutral net of previous step design is trained, tune ginseng is trained using Caffe deep learning frameworks, due to be retrieved
Image there is no label, it is impossible to be trained using data to be retrieved, so using the data set of disclosed tape label to network
Model is trained, and the training dataset that the present invention is used is the data in ImageNet.
Offline forward calculation (forWord) is carried out to the icon in full storehouse using the convolutional neural networks trained, then
Record the 8th layer of value.
By above step, it is all 0 that every image, which can obtain every one-dimensional value in the characteristic vector of one 64 dimension, vector,
Floating number between to 1, this feature vector is preserved.
Further, step (1) also includes carrying out sampling feature vectors hashed processing, and step (2) also includes to mesh
Mark characteristic vector and carry out hashed processing.
Further, the specific implementation of hashed processing is：A threshold value is set, characteristic vector x is mapped as two
System vector f (x),The characteristic vector is any in sampling feature vectors and target feature vector
One.
Specifically, the recall precision for the characteristic vector progress similarity of 64 dimensions can be very low, present invention employs right
The method that characteristic vector carries out hashed, specific practice is one threshold value of setting, and the threshold value in invention is 0.5, then by feature
Vector x is mapped as 01 binary vector, using equation below：
By above method, the characteristic vector of 64 dimensions becomes the binary vector of 64 dimensions, the binary vector of 64 dimension
It is used as final characteristic vector.
According to the introduction to above local sensitivity hash algorithm, the search problem in the present invention is also converted into higher dimensional space
In similarity problem, in order to solve this problem, present invention employs the core concept of LSH algorithms, by feature according to not
Same hash function is put into different buckets, then in retrieval, is retrieved in identical bucket.The present invention is according to LSH algorithms
Thought, feature is uniformly divided into four parts, and keyl, key2, key3, key4 are regarded in this four parts, and keyl is
0101010111000101, key2 is that 0001100110000111, key3 is that 0010100110001101, key4 is
0110001100011100, each feature is the binary vector of one 16 dimension, is then converted to these binary vectors
One hexadecimal number, as last id, such Feature Conversion is into 4 id：Idl, id2, id3, id4, id1 are 55c5,
Id2 is that 1987, id3 is 298d, and id4 is 631c, and each id is a hexadecimal number, as shown in Figure 3.
After handling feature, feature is stored in another search system, in the present invention, employed
ElasticSearch search systems, new index (index) is set up in ES systems, then each feature is regard as files-designated
Know in document (document) deposit system, the feature included in file identification document has：Id1, id2, id3, id4, feature to
Measure the attributes such as value, file MD5 after hashed processing.
With APP growth, the search column provided for the requirement more and more higher of APP retrieval technique, usual APP shops
All it is that corresponding APP is searched by keyword, inputs APP keyword, searches in the description information for then removing APP corresponding
APP, this generally requires to input substantial amounts of description information when by APP typings, and these certain information can also be defeated by developer
Enter.But, for the worker for Safety Industry, the description information only according to APP to carry out safety analysis to APP,
A difficult thing, especially for the APP counterfeit to some detection, counterfeit APP description informations may with it is true
APP it is different or completely uncorrelated, but counterfeit APP is much like with real APP on icon, from the angle of user
Degree is difficult to distinguish the difference of both, and in order to distinguish these counterfeit APP, a feasible method is the icon according to the APP
Removal search goes out similar APP, and the then set further according to these similar APP goes to determine that the APP is counterfeit APP.From another
Consider in the angle of individual message analysis, the search based on APP icons is also necessary, such as the APP of certain bank should
With by the counterfeit personation of malice, bank during how many counterfeit APP, is gone in order to which which counterfeiter excavated out by icon
Scan for, find out similar APP, then filter out benign APP, it is remaining just to find by some counterfeit APP, enter
One step can search the information of counterfeit APP developer, and then investigate and affix legal liability.
Accuracy rate is low in being retrieved instant invention overcomes traditional images, and the problem of recall precision is not high provides the user simultaneously
Very big convenience, can search for corresponding application program of mobile phone according to icon；Simultaneously also for safety detection personnel for counterfeit
APP detection provides a new approach；What is more important is that counterfeit APP or issue are hit by bank or some mechanisms
Safety warning etc. provides technical support.
As it will be easily appreciated by one skilled in the art that the foregoing is merely illustrative of the preferred embodiments of the present invention, it is not used to
The limitation present invention, any modifications, equivalent substitutions and improvements made within the spirit and principles of the invention etc., it all should include
Within protection scope of the present invention.
Claims (10)
1. a kind of similar icon search methods of APP based on convolutional neural networks, it is characterised in that including：
(1) for sample APP icons, sampling feature vectors are extracted based on convolutional neural networks, the sample of sampling feature vectors is recorded
This document is identified, and sampling feature vectors are divided into N number of sample portion, an index is set up in each sample portion, obtains N number of
Index, by sampling feature vectors, sample file mark and N number of index deposit searching system；
(2) for target APP icons, target feature vector is extracted based on convolutional neural networks, target feature vector is divided into N
Individual target part, is combined to each target part, obtains M objective cross characteristic vector；
(3) to M assemblage characteristic vector, M retrieval is carried out in searching system, M file identification set is obtained, by M text
Part identification sets conjunction union, and concentrate and linearly calculated, obtain the similarity of sample APP icons and target APP icons, profit
Sample APP icons are ranked up with similarity.
2. a kind of similar icon search methods of APP based on convolutional neural networks as claimed in claim 1, it is characterised in that
The step (1) also includes pre-processing sample APP icons, and the step (2) also includes carrying out in advance target APP icons
Processing.
3. a kind of similar icon search methods of APP based on convolutional neural networks as claimed in claim 2, it is characterised in that
The pretreatment includes：Image gray processing, image zooming and image normalization.
4. a kind of similar icon search methods of APP based on convolutional neural networks as claimed in claim 1, it is characterised in that
The step (1) also includes carrying out sampling feature vectors hashed processing, and the step (2) also includes to target feature vector
Carry out hashed processing.
5. a kind of similar icon search methods of APP based on convolutional neural networks as claimed in claim 4, it is characterised in that
The specific implementation of hashed processing is：A threshold value is set, characteristic vector x is mapped as binary vector f (x),The characteristic vector is any one in sampling feature vectors and target feature vector.
6. a kind of similar icon searching systems of APP based on convolutional neural networks, it is characterised in that including：
Searching system module, for for sample APP icons, sampling feature vectors to be extracted based on convolutional neural networks, records sample
Sampling feature vectors are divided into N number of sample portion, one are set up in each sample portion by the sample file mark of eigen vector
Individual index, obtains N number of index, by sampling feature vectors, sample file mark and N number of index deposit searching system；
Assemblage characteristic vector module, will for for target APP icons, target feature vector to be extracted based on convolutional neural networks
Target feature vector is divided into N number of target part, and each target part is combined, and obtains M objective cross characteristic vector；
Icon order module, for M assemblage characteristic vector, M retrieval being carried out in searching system, M files-designated is obtained
Know set, by M file identification collection conjunction union, and concentrate and linearly calculated, obtain sample APP icons and target APP
The similarity of icon, is ranked up using similarity to sample APP icons.
7. a kind of similar icon searching systems of APP based on convolutional neural networks as claimed in claim 6, it is characterised in that
The searching system module also includes pre-processing sample APP icons, and the assemblage characteristic vector module also includes to mesh
Mark APP icons are pre-processed.
8. a kind of similar icon searching systems of APP based on convolutional neural networks as claimed in claim 7, it is characterised in that
The pretreatment includes：Image gray processing, image zooming and image normalization.
9. a kind of similar icon searching systems of APP based on convolutional neural networks as claimed in claim 6, it is characterised in that
The searching system module also includes carrying out hashed processing to sampling feature vectors, and the assemblage characteristic vector module also includes
Hashed processing is carried out to target feature vector.
10. a kind of similar icon searching systems of APP based on convolutional neural networks as claimed in claim 9, it is characterised in that
The specific implementation of hashed processing is：A threshold value is set, characteristic vector x is mapped as binary vector f (x),The characteristic vector is any one in sampling feature vectors and target feature vector.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710366244.5A CN107220325A (en) | 2017-05-22 | 2017-05-22 | A kind of similar icon search methods of APP based on convolutional neural networks and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710366244.5A CN107220325A (en) | 2017-05-22 | 2017-05-22 | A kind of similar icon search methods of APP based on convolutional neural networks and system |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107220325A true CN107220325A (en) | 2017-09-29 |
Family
ID=59944300
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710366244.5A Pending CN107220325A (en) | 2017-05-22 | 2017-05-22 | A kind of similar icon search methods of APP based on convolutional neural networks and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107220325A (en) |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108416212A (en) * | 2018-03-01 | 2018-08-17 | 腾讯科技（深圳）有限公司 | Method for identifying application program and device |
| CN111026922A (en) * | 2019-12-26 | 2020-04-17 | 长城计算机软件与系统有限公司 | Distributed vector indexing method, system, plug-in and electronic equipment |
| CN111460185A (en) * | 2020-03-30 | 2020-07-28 | 小船出海教育科技（北京）有限公司 | Book searching method, device and system |
| CN111611264A (en) * | 2020-05-21 | 2020-09-01 | 济南欣格信息科技有限公司 | APP icon retrieval system |
| CN112100046A (en) * | 2020-09-01 | 2020-12-18 | 广州掌淘网络科技有限公司 | Method and equipment for identifying game similarity |
| CN112364309A (en) * | 2021-01-13 | 2021-02-12 | 北京云真信科技有限公司 | Information processing method, electronic device, and computer-readable storage medium |
| CN112487437A (en) * | 2020-11-17 | 2021-03-12 | 北京航空航天大学 | Android counterfeit application detection method and device based on user interface characteristics |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105095435A (en) * | 2015-07-23 | 2015-11-25 | 北京京东尚科信息技术有限公司 | Similarity comparison method and device for high-dimensional image features |
| CN105512289A (en) * | 2015-12-07 | 2016-04-20 | 郑州金惠计算机系统工程有限公司 | Image retrieval method based on deep learning and Hash |
| US20160357748A1 (en) * | 2015-06-04 | 2016-12-08 | Yahoo!, Inc. | Image searching |
| US20160358024A1 (en) * | 2015-06-03 | 2016-12-08 | Hyperverge Inc. | Systems and methods for image processing |
| CN106227851A (en) * | 2016-07-29 | 2016-12-14 | 汤平 | Based on the image search method searched for by depth of seam division that degree of depth convolutional neural networks is end-to-end |
| CN106682233A (en) * | 2017-01-16 | 2017-05-17 | 华侨大学 | Method for Hash image retrieval based on deep learning and local feature fusion |
-
2017
- 2017-05-22 CN CN201710366244.5A patent/CN107220325A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20160358024A1 (en) * | 2015-06-03 | 2016-12-08 | Hyperverge Inc. | Systems and methods for image processing |
| US20160357748A1 (en) * | 2015-06-04 | 2016-12-08 | Yahoo!, Inc. | Image searching |
| CN105095435A (en) * | 2015-07-23 | 2015-11-25 | 北京京东尚科信息技术有限公司 | Similarity comparison method and device for high-dimensional image features |
| CN105512289A (en) * | 2015-12-07 | 2016-04-20 | 郑州金惠计算机系统工程有限公司 | Image retrieval method based on deep learning and Hash |
| CN106227851A (en) * | 2016-07-29 | 2016-12-14 | 汤平 | Based on the image search method searched for by depth of seam division that degree of depth convolutional neural networks is end-to-end |
| CN106682233A (en) * | 2017-01-16 | 2017-05-17 | 华侨大学 | Method for Hash image retrieval based on deep learning and local feature fusion |
Non-Patent Citations (3)
| Title |
|---|
| 沈庭芝等: "《数字图像处理及模式识别 第2版》", 31 July 2007, 北京理工大学出版社 * |
| 鱼滨等: "《基于MATLAB和遗传算法的图像处理》", 31 October 2015, 西安电子科技大学出版社 * |
| 黄惠芬等: "《数字图像司法取证技术》", 30 April 2015, 山东大学出版社 * |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108416212A (en) * | 2018-03-01 | 2018-08-17 | 腾讯科技（深圳）有限公司 | Method for identifying application program and device |
| CN111026922A (en) * | 2019-12-26 | 2020-04-17 | 长城计算机软件与系统有限公司 | Distributed vector indexing method, system, plug-in and electronic equipment |
| CN111026922B (en) * | 2019-12-26 | 2024-05-28 | 新长城科技有限公司 | Distributed vector indexing method, system, plug-in and electronic equipment |
| CN111460185A (en) * | 2020-03-30 | 2020-07-28 | 小船出海教育科技（北京）有限公司 | Book searching method, device and system |
| CN111611264A (en) * | 2020-05-21 | 2020-09-01 | 济南欣格信息科技有限公司 | APP icon retrieval system |
| CN111611264B (en) * | 2020-05-21 | 2022-06-10 | 济南欣格信息科技有限公司 | APP icon retrieval system |
| CN112100046A (en) * | 2020-09-01 | 2020-12-18 | 广州掌淘网络科技有限公司 | Method and equipment for identifying game similarity |
| CN112487437A (en) * | 2020-11-17 | 2021-03-12 | 北京航空航天大学 | Android counterfeit application detection method and device based on user interface characteristics |
| CN112364309A (en) * | 2021-01-13 | 2021-02-12 | 北京云真信科技有限公司 | Information processing method, electronic device, and computer-readable storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107220325A (en) | A kind of similar icon search methods of APP based on convolutional neural networks and system | |
| US10867206B2 (en) | Content based search and retrieval of trademark images | |
| CN109815952A (en) | Brand name recognition methods, computer installation and computer readable storage medium | |
| US20170024384A1 (en) | System and method for analyzing and searching imagery | |
| CN104573130B (en) | The entity resolution method and device calculated based on colony | |
| CN104537341B (en) | Face picture information getting method and device | |
| CN105159938B (en) | Search method and device | |
| Srivastava et al. | Content-based image retrieval using moments of local ternary pattern | |
| CN112884005B (en) | Image retrieval method and device based on SPTAG and convolutional neural network | |
| US20190197133A1 (en) | Shape-based graphics search | |
| CN111125396B (en) | Image retrieval method of single-model multi-branch structure | |
| CN111767420B (en) | Method and device for generating clothing collocation data | |
| CN103838754A (en) | Information searching device and method | |
| CN104317946A (en) | Multi-key image-based image content retrieval method | |
| CN105183857A (en) | Automatic picture training sample extracting method and system | |
| Zhai et al. | FMT: fusing multi-task convolutional neural network for person search | |
| CN108121806A (en) | One kind is based on the matched image search method of local feature and system | |
| CN109697240A (en) | A kind of image search method and device based on feature | |
| CN109241315B (en) | Rapid face retrieval method based on deep learning | |
| CN105069003B (en) | A kind of user's perpetual object based on forwarding chain similarity recommends computational methods | |
| CN117332103A (en) | Image retrieval method based on keyword extraction and multi-modal feature fusion | |
| CN110377790B (en) | Video automatic labeling method based on multi-mode private features | |
| Hezel et al. | Video search with sub-image keyword transfer using existing image archives | |
| CN111159411A (en) | Knowledge graph fused text position analysis method, system and storage medium | |
| Zhang et al. | Embedding multi-order spatial clues for scalable visual matching and retrieval |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20170929 |
|
| RJ01 | Rejection of invention patent application after publication |