CN108154182A - Jewelry method for measuring similarity based on pairs of comparing cell - Google Patents
Jewelry method for measuring similarity based on pairs of comparing cell Download PDFInfo
- Publication number
- CN108154182A CN108154182A CN201711420434.7A CN201711420434A CN108154182A CN 108154182 A CN108154182 A CN 108154182A CN 201711420434 A CN201711420434 A CN 201711420434A CN 108154182 A CN108154182 A CN 108154182A
- Authority
- CN
- China
- Prior art keywords
- pairs
- comparing cell
- similarity
- jewelry
- measuring similarity
- 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.)
- Withdrawn
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- 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
- G06N3/048—Activation functions
Abstract
The invention discloses the jewelry method for measuring similarity based on pairs of comparing cell, are related to picture comparison technology field.The present invention includes the following steps：A, construction compares the training dataset of network in pairs, B, the feature extraction network resnet50 of several parameter sharings is established, C, the resnet50 characteristic patterns extracted are spliced together and are mapped with the similarity between convolutional layer learns, and activation primitive uses Sigmoid functions, D, a pair of of image is inputted, by obtaining the similarity between 0 to 1 after Double Net.The present invention opens the similitude between figure by neural network learning two, extract the feature of two figures respectively using CNN, then it is being (1 with convolution kernel, 1), the decision-making level for 1 neuron is exported to learn the mapping relations of metric function, method all by convolutional layer operation, can be arbitrary size to input tomographic image, improve relative efficiency.
Description
Technical field
The invention belongs to picture comparison technology field, more particularly to a kind of jewelry similitude based on pairs of comparing cell
Measure.
Background technology
The similitude that traditional method weighs between image is that extract the geometric properties such as the texture of image, shape, color right
After be spliced into a vector, the similarity between two images is exactly with profound, Pierre more than Euclidean distance, manhatton distance, angle
The inferior isocratic flow function of related coefficient is weighed.It is inaccurate that this method is easy to cause measurement.
Convolutional neural networks are a kind of feedforward neural networks, and artificial neuron can respond surrounding cells, can be with Large Graph
As processing.Convolutional neural networks include convolutional layer and pond layer.Usually, the basic structure of CNN includes two layers, and one is characterized
Extract layer, the input of each neuron are connected, and extract the feature of the part with the local acceptance region of preceding layer.Once the part
After feature is extracted, its position relationship between other feature is also decided therewith；The second is Feature Mapping layer, network it is every
A computation layer is made of multiple Feature Mappings, and each Feature Mapping is a plane, and the weights of all neurons are equal in plane.
Feature Mapping structure is using activation primitive of the small Sigmoid functions of influence function core as convolutional network so that Feature Mapping
With shift invariant.Further, since the neuron on a mapping face shares weights, thus reduce network freedom parameter
Number.Each convolutional layer followed by one in convolutional neural networks is used for asking the calculating of local average and second extraction
Layer, this distinctive structure of feature extraction twice reduce feature resolution.CNN be mainly used to identify displacement, scaling and other
Form distorts the X-Y scheme of invariance.
The development of random deep learning and artificial intelligence technology, by constantly improve in Application of Neural Network to image.
In pervious technology, generally use is the feature that image is extracted with convolutional neural networks, is then weighed again with metric function
Measure similitude.This method is not compared end to end, is easy to cause measurement and is failed, in the present invention using with god
Learn the similitude between two images through network.
Invention content
The purpose of the present invention is to provide a kind of jewelry method for measuring similarity based on pairs of comparing cell, pass through nerve
E-learning two opens the similitude between figure, extracts the feature of two figures respectively using CNN, and it is (1,1) then to use convolution kernel again,
The decision-making level for 1 neuron is exported to learn the mapping relations of metric function, solves and convolutional Neural is first used in traditional technology
Network extracts the feature of image, then weighs similitude with metric function, is easy to cause the problem of measurement fails.
In order to solve the above technical problems, the present invention is achieved by the following technical solutions：
The present invention is the jewelry method for measuring similarity based on pairs of comparing cell, is included the following steps：
Step SS001 constructions compare the training dataset of network in pairs；
Step SS002 establishes the feature extraction network resnet50 of several parameter sharings；
The resnet50 characteristic patterns extracted are spliced together by step SS003 to be reflected with the similarity between convolutional layer learns
It penetrates, and activation primitive uses Sigmoid functions；
A pair of of image of step SS004 inputs, by obtaining the similarity between 0 to 1 after Double Net.
Preferably, in the step SS001, training dataset includes construction training set, construction verification collection and construction test
Collection.
Preferably, in the step SS003, convolutional layer (kernel=(2,2), output=1) is swashed with Sigmoid functions
It is living, wherein, Sigmoid functions can be expressed as：
Preferably, in the step SS004, metric is between 0 to 1；Wherein, similar value is closer to 1, then it represents that two
Picture is more similar.
The invention has the advantages that：
The present invention opens the similitude between figure by neural network learning two, extracts the feature of two figures respectively using CNN,
Then it is being (1,1) with convolution kernel, is exporting the decision-making level for 1 neuron to learn the mapping relations of metric function, method is complete
Portion can be arbitrary size to input tomographic image, improve relative efficiency by convolutional layer operation.
Certainly, it implements any of the products of the present invention and does not necessarily require achieving all the advantages described above at the same time.
Description of the drawings
In order to illustrate the technical solution of the embodiments of the present invention more clearly, embodiment will be described below required
Attached drawing is briefly described, it should be apparent that, the accompanying drawings in the following description is only some embodiments of the present invention, for ability
For the those of ordinary skill of domain, without creative efforts, it can also be obtained according to these attached drawings other attached
Figure.
The step of Fig. 1 is the jewelry method for measuring similarity based on pairs of comparing cell of the present invention is schemed；
Fig. 2 is the flow chart of the jewelry method for measuring similarity based on pairs of comparing cell of the present invention；
Fig. 3 is the application effect figure of the present invention.
Specific embodiment
Below in conjunction with the attached drawing in the embodiment of the present invention, the technical solution in the embodiment of the present invention is carried out clear, complete
Site preparation describes, it is clear that described embodiment is only part of the embodiment of the present invention, instead of all the embodiments.It is based on
Embodiment in the present invention, those of ordinary skill in the art are obtained all other without creative efforts
Embodiment shall fall within the protection scope of the present invention.
Refering to Figure 1, the present invention is the jewelry method for measuring similarity based on pairs of comparing cell, including walking as follows
Suddenly：
Step SS001 constructions compare the training dataset of network, similar data tagged 1, dissimilar number in pairs
According to tagged 0；
Feature extraction the network resnet50, resnet that step SS002 establishes two parameter sharings are current most fiery volume
Product neural network structure, learns to form residual error function by doing each layer one reference, easily optimizes, and increase
The network number of plies；
The resnet50 characteristic patterns extracted are spliced together by step SS003 to be reflected with the similarity between convolutional layer learns
It penetrates, and activation primitive uses Sigmoid functions；
A pair of of image of step SS004 inputs, by obtaining the similarity between 0 to 1 after Double Net.
It please referring to shown in Fig. 2, the present invention is the jewelry method for measuring similarity based on pairs of comparing cell, wherein, step
In SS001, training dataset includes construction training set, construction verification collection and construction test set；Because the jewellery type of jewelry can
To be divided into 5 big classifications：Necklace, bangle, bracelet, ear pendant and thimble, and their quantity is respectively n1, n2, n3, n4 and n5, is incited somebody to action
This 5 class carries out 10 deciles respectively according to collection, and the data per decile are n=max (n1/10, n2/10, n3/10, n4/10, n5/
10), when the data of a certain decile of some classification are inadequate, the direct random total amount of data in the category is selected at random；It will be every
Basic data collection of a kind of wherein 8 grades point as construction training set, basic data collection of 1 grade point as construction verification collection, 1 decile
Basic data collection as construction test set.It is random to select k1 as representative element for each decile, respectively with inhomogeneity
Other n groups pair, label are 0, and the n groups pair with the same category, label is 1.
Wherein, in step SS003, convolutional layer (kernel=(2,2), output=1) is activated with Sigmoid functions,
In, Sigmoid functions can be expressed as：
Sigmoid functions are the function of a common S type in biology, also referred to as S sigmoid growth curves.In Information Center
In, due to it, singly properties, the Sigmoid functions such as increasing and the increasing of inverse function list are often used as the threshold function table of neural network, will
To between 0,1, Sigmoid functions have the following properties that variable mappings：When x levels off to it is negative infinite when, y levels off to 0；When x is approached
When just infinite, y levels off to 1；As x=1/2, y=0.
Wherein, in step SS004, metric is between 0 to 1；Wherein, similar value is closer to 1, then it represents that two pictures are got over
It is similar.
It please refers to shown in Fig. 3, the present invention is the application effect of the jewelry method for measuring similarity based on pairs of comparing cell
Figure, the quantity of training set is 3061125 pairs, and verification collection is 204075 pairs, and test set is 204075 pairs.New a pair of of image input
Into network, it is 1 that the label to image is directly returned when the value of neural network forecast is more than 0.5, otherwise returns to 0.Test set leads to
It is 0.968 to cross accuracy of the neural network forecast compared with the label manually beaten, and the accuracy for verifying collection is 0.945, and test set is just
True rate be 0.923. when inputting lower pair of image, 0.9878 during the return value of network.
It is worth noting that, in above system embodiment, included each unit is only drawn according to function logic
Point, but above-mentioned division is not limited to, as long as corresponding function can be realized；In addition, each functional unit is specific
Title is also only to facilitate mutually distinguish, the protection domain being not intended to restrict the invention.
In addition, one of ordinary skill in the art will appreciate that realize all or part of step in the various embodiments described above method
It is that relevant hardware can be instructed to complete by program, corresponding program can be stored in a computer-readable storage and be situated between
In matter, the storage medium, such as ROM/RAM, disk or CD.
Present invention disclosed above preferred embodiment is only intended to help to illustrate the present invention.There is no detailed for preferred embodiment
All details are described, are not limited the invention to the specific embodiments described.Obviously, according to the content of this specification,
It can make many modifications and variations.This specification is chosen and specifically describes these embodiments, is in order to preferably explain the present invention
Principle and practical application, so as to which skilled artisan be enable to be best understood by and utilize the present invention.The present invention is only
It is limited by claims and its four corner and equivalent.
Claims (4)
1. the jewelry method for measuring similarity based on pairs of comparing cell, which is characterized in that include the following steps：
Step SS001 constructions compare the training dataset of network in pairs；
Step SS002 establishes the feature extraction network resnet50 of several parameter sharings；
The resnet50 characteristic patterns extracted are spliced together by step SS003 to be mapped with the similarity between convolutional layer learns, and
And activation primitive uses Sigmoid functions；
A pair of of image of step SS004 inputs, by obtaining the similarity between 0 to 1 after Double Net.
2. the jewelry method for measuring similarity according to claim 1 based on pairs of comparing cell, which is characterized in that described
In step SS001, training dataset includes construction training set, construction verification collection and construction test set.
3. the jewelry method for measuring similarity according to claim 1 based on pairs of comparing cell, which is characterized in that described
In step SS003, convolutional layer (kernel=(2,2), output=1) is activated with Sigmoid functions, wherein, Sigmoid functions
It can be expressed as：
4. the jewelry method for measuring similarity according to claim 1 based on pairs of comparing cell, which is characterized in that described
In step SS004, metric is between 0 to 1；Wherein, similar value is closer to 1, then it represents that two pictures are more similar.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711420434.7A CN108154182A (en) | 2017-12-25 | 2017-12-25 | Jewelry method for measuring similarity based on pairs of comparing cell |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711420434.7A CN108154182A (en) | 2017-12-25 | 2017-12-25 | Jewelry method for measuring similarity based on pairs of comparing cell |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108154182A true CN108154182A (en) | 2018-06-12 |
Family
ID=62465794
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201711420434.7A Withdrawn CN108154182A (en) | 2017-12-25 | 2017-12-25 | Jewelry method for measuring similarity based on pairs of comparing cell |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108154182A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109376741A (en) * | 2018-09-10 | 2019-02-22 | 平安科技（深圳）有限公司 | Recognition methods, device, computer equipment and the storage medium of trademark infringement |
| WO2020232941A1 (en) * | 2019-05-17 | 2020-11-26 | 丰疆智能科技股份有限公司 | Dairy cattle nipple detection convolutional neural network model and construction method therefor |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20070172155A1 (en) * | 2006-01-21 | 2007-07-26 | Elizabeth Guckenberger | Photo Automatic Linking System and method for accessing, linking, and visualizing "key-face" and/or multiple similar facial images along with associated electronic data via a facial image recognition search engine |
| CN105574510A (en) * | 2015-12-18 | 2016-05-11 | 北京邮电大学 | Gait identification method and device |
| CN106022380A (en) * | 2016-05-25 | 2016-10-12 | 中国科学院自动化研究所 | Individual identity identification method based on deep learning |
| CN106503106A (en) * | 2016-10-17 | 2017-03-15 | 北京工业大学 | A kind of image hash index construction method based on deep learning |
| CN107292259A (en) * | 2017-06-15 | 2017-10-24 | 国家新闻出版广电总局广播科学研究院 | The integrated approach of depth characteristic and traditional characteristic based on AdaRank |
-
2017
- 2017-12-25 CN CN201711420434.7A patent/CN108154182A/en not_active Withdrawn
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20070172155A1 (en) * | 2006-01-21 | 2007-07-26 | Elizabeth Guckenberger | Photo Automatic Linking System and method for accessing, linking, and visualizing "key-face" and/or multiple similar facial images along with associated electronic data via a facial image recognition search engine |
| CN105574510A (en) * | 2015-12-18 | 2016-05-11 | 北京邮电大学 | Gait identification method and device |
| CN106022380A (en) * | 2016-05-25 | 2016-10-12 | 中国科学院自动化研究所 | Individual identity identification method based on deep learning |
| CN106503106A (en) * | 2016-10-17 | 2017-03-15 | 北京工业大学 | A kind of image hash index construction method based on deep learning |
| CN107292259A (en) * | 2017-06-15 | 2017-10-24 | 国家新闻出版广电总局广播科学研究院 | The integrated approach of depth characteristic and traditional characteristic based on AdaRank |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109376741A (en) * | 2018-09-10 | 2019-02-22 | 平安科技（深圳）有限公司 | Recognition methods, device, computer equipment and the storage medium of trademark infringement |
| WO2020232941A1 (en) * | 2019-05-17 | 2020-11-26 | 丰疆智能科技股份有限公司 | Dairy cattle nipple detection convolutional neural network model and construction method therefor |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US11417148B2 (en) | Human face image classification method and apparatus, and server | |
| CN108108807B (en) | Learning type image processing method, system and server | |
| CN106095893B (en) | A kind of cross-media retrieval method | |
| Lake et al. | Deep neural networks predict category typicality ratings for images. | |
| CN107818314B (en) | Face image processing method, device and server | |
| US11526765B2 (en) | Systems and methods for a supra-fusion graph attention model for multi-layered embeddings and deep learning applications | |
| CN105046277B (en) | Robust mechanism study method of the feature significance in image quality evaluation | |
| CN110046698A (en) | Heterogeneous figure neural network generation method, device, electronic equipment and storage medium | |
| Bodhwani et al. | Deep residual networks for plant identification | |
| Zhang et al. | Plant species recognition methods using leaf image: Overview | |
| CN110097003A (en) | Check class attendance method, equipment, storage medium and device neural network based | |
| Pan et al. | A central-point-enhanced convolutional neural network for high-resolution remote-sensing image classification | |
| CN104298682A (en) | Information recommendation effect evaluation method and mobile phone based on facial expression images | |
| CN110781897A (en) | Semantic edge detection method based on deep learning | |
| Benzebouchi et al. | Multi-modal classifier fusion with feature cooperation for glaucoma diagnosis | |
| US11093800B2 (en) | Method and device for identifying object and computer readable storage medium | |
| CN116933046B (en) | Deep learning-based multi-mode health management scheme generation method and system | |
| Morgenstern et al. | An image-computable model of human visual shape similarity | |
| CN107341440A (en) | Indoor RGB D scene image recognition methods based on multitask measurement Multiple Kernel Learning | |
| CN113705596A (en) | Image recognition method and device, computer equipment and storage medium | |
| CN108154182A (en) | Jewelry method for measuring similarity based on pairs of comparing cell | |
| Wagner et al. | Deep learning and adaptive graph-based growing contours for agricultural field extraction | |
| CN110363072A (en) | Tongue image recognition method, apparatus, computer equipment and computer readable storage medium | |
| CN110378237A (en) | Human facial expression recognition method based on depth measure converged network | |
| CN108985442A (en) | Handwriting model training method, hand-written character recognizing method, device, equipment and medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| TA01 | Transfer of patent application right | ||
| TA01 | Transfer of patent application right |
Effective date of registration: 20210108 Address after: Room 902, 9 / F, building F, Tianhe Software Park, Tianhe District, Guangzhou 510000 Applicant after: GUANGDONG 3VJIA INFORMATION TECHNOLOGY Co.,Ltd. Address before: 230000 room 323, building E1, phase II, innovation industrial park, 2800 innovation Avenue, high tech Zone, Hefei City, Anhui Province Applicant before: HEFEI ABACI SCIENCE & TECHNOLOGY Co.,Ltd. |
|
| WW01 | Invention patent application withdrawn after publication | ||
| WW01 | Invention patent application withdrawn after publication |
Application publication date: 20180612 |