CN108921181A - A kind of local image characteristics extracting method, device, system and readable storage medium storing program for executing - Google Patents
A kind of local image characteristics extracting method, device, system and readable storage medium storing program for executing Download PDFInfo
- Publication number
- CN108921181A CN108921181A CN201810870557.9A CN201810870557A CN108921181A CN 108921181 A CN108921181 A CN 108921181A CN 201810870557 A CN201810870557 A CN 201810870557A CN 108921181 A CN108921181 A CN 108921181A
- Authority
- CN
- China
- Prior art keywords
- topography
- similarity
- deconvolution parameter
- 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.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 64
- 238000003860 storage Methods 0.000 title claims abstract description 20
- 238000012876 topography Methods 0.000 claims abstract description 75
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 47
- 238000000605 extraction Methods 0.000 claims abstract description 38
- 238000002715 modification method Methods 0.000 claims abstract description 6
- 230000006870 function Effects 0.000 claims description 27
- 238000004364 calculation method Methods 0.000 claims description 13
- 238000004590 computer program Methods 0.000 claims description 6
- 239000000284 extract Substances 0.000 claims description 5
- 238000011478 gradient descent method Methods 0.000 claims description 3
- 230000000694 effects Effects 0.000 abstract description 7
- 230000009286 beneficial effect Effects 0.000 abstract description 4
- 230000008569 process Effects 0.000 description 12
- 238000010586 diagram Methods 0.000 description 10
- 238000010606 normalization Methods 0.000 description 7
- 230000004913 activation Effects 0.000 description 6
- 238000004422 calculation algorithm Methods 0.000 description 4
- 238000010168 coupling process Methods 0.000 description 3
- 238000005859 coupling reaction Methods 0.000 description 3
- 230000004069 differentiation Effects 0.000 description 3
- 230000001537 neural effect Effects 0.000 description 3
- 230000008439 repair process Effects 0.000 description 3
- 238000012549 training Methods 0.000 description 3
- 238000013528 artificial neural network Methods 0.000 description 2
- 238000004891 communication Methods 0.000 description 2
- 230000008878 coupling Effects 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 241000208340 Araliaceae Species 0.000 description 1
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 description 1
- 235000003140 Panax quinquefolius Nutrition 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000005520 cutting process Methods 0.000 description 1
- 238000005034 decoration Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 230000008034 disappearance Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000001815 facial effect Effects 0.000 description 1
- 230000014509 gene expression Effects 0.000 description 1
- 235000008434 ginseng Nutrition 0.000 description 1
- 238000005286 illumination Methods 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 230000002779 inactivation Effects 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000013178 mathematical model Methods 0.000 description 1
- 230000005012 migration Effects 0.000 description 1
- 238000013508 migration Methods 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000005192 partition Methods 0.000 description 1
- 239000011148 porous material Substances 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 238000011002 quantification Methods 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000000926 separation method Methods 0.000 description 1
- 238000013519 translation Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
This application discloses a kind of local image characteristics extracting methods, including：Deconvolution parameter in convolutional neural networks is modified, revised deconvolution parameter is obtained；Obtain topography；The characteristic information for extracting topography according to revised deconvolution parameter based on convolutional neural networks, obtains feature vector；Wherein, the modification method of deconvolution parameter includes：What be will acquire includes different classes of topography's information input to convolutional neural networks, obtains the corresponding feature vector of each topography's information；Calculate the corresponding characteristic central point of topography's information of all categories；The similarity of topography in of all categories and same category is calculated according to characteristic central point；Deconvolution parameter is modified according to similarity.The extracting method local shape factor ability is strong, and Minutiae extraction effect is good.Disclosed herein as well is local image characteristics extraction element, system and computer readable storage mediums, have above-mentioned beneficial effect.
Description
Technical field
This application involves domain-adaptive field, in particular to a kind of local image characteristics extracting method, device, system and
A kind of computer readable storage medium.
Background technique
Local image characteristics are the local expressions of characteristics of image, it reflects the local characteristics having on image, with line spy
The global images feature such as sign, textural characteristics, structure feature is compared, and local image characteristics are abundant with quantity is contained in the picture,
The degree of correlation is small between feature, under circumstance of occlusion will not because of Partial Feature disappearance and detection and the matching etc. that influence other features is excellent
Gesture has obtained widely in the fields such as recognition of face, three-dimensional reconstruction, target identification and tracking, production of film and TV, Panorama Mosaic
Using.The extraction of local image characteristics typically as many problems in computer vision and Digital Image Processing the first step,
Such as image classification, image retrieval, wide baseline matching etc., the superiority and inferiority for extracting feature directly affects the final performance of task.Part
Characteristics of image description is a basic research problem of computer vision, and the corresponding points and object features in searching image are retouched
There is important role in stating.Therefore, Local Feature Extraction has important researching value.
However, image scale, translation, rotation, illumination, visual angle and the variation such as fuzzy frequent occurrence, especially in reality
In application scenarios, inevitably there is larger noise jamming, complex background and biggish target carriage change in image, at this
The extraction effect of local feature tends not to fully up to expectations under kind rough sledding.
Therefore, local shape factor ability how is promoted, Minutiae extraction effect is promoted, is that those skilled in the art need
Technical problems to be solved.
Summary of the invention
The purpose of the application is to provide a kind of local image characteristics extracting method, the extracting method local shape factor ability
By force, Minutiae extraction effect is good；The another object of the application is to provide a kind of local image characteristics extraction element, system and one
Kind computer readable storage medium, has above-mentioned beneficial effect.
The application provides a kind of local image characteristics extracting method, including：
Deconvolution parameter in convolutional neural networks is modified, revised deconvolution parameter is obtained；
Obtain topography；
Believed based on the convolutional neural networks according to the feature that the revised deconvolution parameter extracts the topography
Breath, obtains feature vector；
Wherein, the modification method of the deconvolution parameter includes：
What be will acquire includes different classes of topography's information input to convolutional neural networks, obtains each topography's letter
Cease corresponding feature vector；
Calculate the corresponding characteristic central point of topography's information of all categories；
The similarity of topography in of all categories and same category is calculated according to the characteristic central point；
The deconvolution parameter is modified according to the similarity.
Optionally, described that the similar of topography in of all categories and same category is calculated according to the characteristic central point
Degree includes：
Calculate the Euclidean distance of each feature vector and corresponding characteristic central point in same category；
Count the fluctuation situation that each feature vector corresponds to Euclidean distance, as in corresponding classification topography it is similar
Degree；
The Euclidean distance between variant classification between characteristic central point is calculated, as the similar of of all categories topography
Degree.
Optionally, the fluctuation situation that each feature vector of the statistics corresponds to Euclidean distance includes：
It calculates each feature vector and corresponds to variance between Euclidean distance.
Optionally, it is described according to the similarity to the deconvolution parameter be modified including：
The loss function of the convolutional neural networks is constructed according to the similarity；
It is modified according to the deconvolution parameter of the loss function to the convolutional neural networks.
Optionally, described that packet is modified according to the deconvolution parameter of the loss function to the convolutional neural networks
It includes：
According to the loss function based on stochastic gradient descent method to the deconvolution parameter of the convolutional neural networks
It is modified.
Optionally, the acquisition topography includes：
Receive general image；
Receive the local message for carrying out feature extraction；
Described image is cut according to the local message, obtains topography.
The application discloses a kind of local image characteristics extraction element, including：
Amending unit obtains revised deconvolution parameter for being modified to the deconvolution parameter in convolutional neural networks；
Acquiring unit, for obtaining topography；
Feature extraction unit extracts the part according to the revised deconvolution parameter based on the convolutional neural networks
The characteristic information of image, obtains feature vector；
Wherein, the amending unit includes：
Image inputs subelement, for will acquire including different classes of topography's information input to convolutional Neural net
Network obtains the corresponding feature vector of each topography's information；
Central point computation subunit, for calculating the corresponding characteristic central point of topography's information of all categories；
Similarity calculation subelement, for calculating part in of all categories and same category according to the characteristic central point
The similarity of image；
Parameters revision subelement, for being modified according to the similarity to the deconvolution parameter.
Optionally, the similarity calculation subelement includes：
First computation subunit, for calculating in each feature vector is several with the Europe of corresponding characteristic central point in same category
Obtain distance；
Subelement is counted, the fluctuation situation of Euclidean distance is corresponded to for counting each feature vector, as corresponding classification
The similarity of interior topography；
Second computation subunit, for calculating the Euclidean distance between variant classification between characteristic central point, as each
The similarity of topography between classification.
The application discloses a kind of local image characteristics extraction system, which is characterized in that including：
Memory, for storing computer program；
Processor, when for executing the computer program the step of local image characteristics extracting method described in realization.
The application discloses a kind of readable storage medium storing program for executing, and program is stored on the readable storage medium storing program for executing, and described program is located
The step of reason device realizes the local image characteristics extracting method when executing.
In order to solve the above technical problems, the application provides a kind of local image characteristics extracting method, this method passes through convolution
Neural network extracts the characteristic information of topography, and convolutional neural networks can be found that and portray knot complicated inside topography
Structure feature, so being capable of the larger performance for improving feature extraction；In addition, defeated by that will include different classes of topography's information
Enter to convolutional neural networks, after obtaining feature vector, to the corresponding feature vector of each topography's information by calculating in feature
The mode of heart point is found in the same category and the similarity degree of different classes of characteristics of image, is assessed and is learnt by similarity degree
Network migrates the differentiation degree in similar image and inhomogeneity image to inhomogeneity image characteristic extracting method,
The extraction of similar local image characteristics is distinguished, deconvolution parameter is constantly corrected, obtains more precisely portraying details
The deconvolution parameter of distinguishing characteristics, makeover process only need to calculate similarity degree, and calculation amount is smaller, and makeover process is rapid, convolution
Convolutional neural networks after parameters revision can more accurately excavate corresponding class another characteristic, and image recognition rate is high, and robustness is high.
Disclosed herein as well is a kind of local image characteristics extraction element, system and a kind of computer readable storage medium,
With above-mentioned beneficial effect, details are not described herein.
Detailed description of the invention
In order to illustrate the technical solutions in the embodiments of the present application or in the prior art more clearly, to embodiment or will show below
There is attached drawing needed in technical description to be briefly described, it should be apparent that, the accompanying drawings in the following description is only this
The embodiment of application for those of ordinary skill in the art without creative efforts, can also basis
The attached drawing of offer obtains other attached drawings.
Fig. 1 is the flow chart of local image characteristics extracting method provided by the embodiments of the present application；
Fig. 2 is the flow chart of the modification method of deconvolution parameter provided by the embodiments of the present application；
Fig. 3 is the flow chart of similarity calculating method provided by the embodiments of the present application；
Fig. 4 is the structural block diagram of local image characteristics extraction element provided by the embodiments of the present application；
Fig. 5 is the structural block diagram of amending unit provided by the embodiments of the present application；
Fig. 6 is similarity calculation sub-unit structure block diagram provided by the embodiments of the present application；
Fig. 7 is the structural block diagram of parameters revision subelement provided by the embodiments of the present application；
Fig. 8 is the structural block diagram of acquiring unit provided by the embodiments of the present application.
Specific embodiment
The core of the application is to provide a kind of text information extracting method based on domain-adaptive, and this method can be improved
Field migration promotes the text analyzing extractability in the fields such as social media；Another core of the application is to provide a kind of base
In the text information extraction element of domain-adaptive, system and a kind of readable storage medium storing program for executing, there is above-mentioned beneficial effect.
To keep the purposes, technical schemes and advantages of the embodiment of the present application clearer, below in conjunction with the embodiment of the present application
In attached drawing, the technical scheme in the embodiment of the application is clearly and completely described, it is clear that described embodiment is
Some embodiments of the present application, instead of all the embodiments.Based on the embodiment in the application, those of ordinary skill in the art
Every other embodiment obtained without making creative work, shall fall in the protection scope of this application.
Fig. 1 is the flow chart of local image characteristics extracting method provided in this embodiment；This method may include：
Step s100：Deconvolution parameter in convolutional neural networks is modified, revised deconvolution parameter is obtained.
Wherein, Fig. 2 is the flow chart of the modification method of deconvolution parameter provided in this embodiment, the modification method of deconvolution parameter
Specifically include following steps：
Step s110：What be will acquire includes different classes of topography's information input to convolutional neural networks, is obtained each
The corresponding feature vector of topography's information；
Any to choose k classification, it is that the part n training image is input to convolution mind that number is respectively randomly selected in each classification
Through in network, it is not limited here, being introduced so that local training image is 32 × 32 pixels as an example herein can for image pixel value
To obtain description of the feature vector as image of total k × n 128 dimension of k group.
Step s120：Calculate the corresponding characteristic central point of topography's information of all categories；
Every group of central point is found out in the k group feature vector obtained in step s110, characteristic central point refers to feature vector
Vector mean value.For example, feature vector is respectively 1,2,3,4,5 in classification 1, then central point is 3.Wherein, characteristic central point
It is not limited only to the vector for including in feature vector.
Step s130：The similarity of topography in of all categories and same category is calculated according to characteristic central point；
The similarity of each feature vector Yu central feature vector is assessed, in every group to judge similar topography
Similarity；The similarity of central point feature vector between two groups is assessed, to judge the similarity of the topography of different classifications.
The portion of reaction image is only capable of by the description that the modes such as imaging angle, gray scale and collection location carry out characteristics of image
Dtex sign, local image characteristics extracting method provided in this embodiment are started with to the comprehensive character of image, are promoted by similarity
To the differentiation degree of characteristics of image, thus find the small differences between image and image, feature extraction side provided in this embodiment
The image recognition of pore rank may be implemented in method, and feature extraction effect is good.
Wherein, calculate the method for similarity without limitation, for example, the similarity in every group can calculate each feature vector with
The sum of difference of central point, different classes of similarity can be by the difference of central point between different groups of calculating, as two
Similarity between group；Similarity etc. can also be assessed by calculating Euclidean distance.Preferably, as far as possible for guarantee algorithm
Vector is accurately assessed in simple situation, similarity can be assessed by calculating Euclidean distance, specifically,
Fig. 3 is the flow chart of similarity calculating method provided in this embodiment, calculates of all categories and same according to characteristic central point
The similarity of topography may include in classification：
Step s131：Calculate the Euclidean distance of each feature vector and corresponding characteristic central point in same category；
Step s132：The fluctuation situation that each feature vector corresponds to Euclidean distance is counted, as part in corresponding classification
The similarity of image；
Step s133：The Euclidean distance between variant classification between characteristic central point is calculated, as of all categories part
The similarity of image.
The specific algorithm of statistical fluctuation situation can refer to the prior art in step s132, it is not limited here, such as can
By calculating variance, standard deviation etc., it is preferable that, can be by calculating each feature vector pair to guarantee calculating process simplification
The variance between Euclidean distance is answered to determine fluctuation situation.
Step s140：Deconvolution parameter is modified according to similarity.
The quality of current convolutional neural networks performance is assessed using similarity information obtained in step s130；It is different classes of
Between similarity degree it is more low more can reflect the corresponding unique feature of each classification, the similarity in the same category the low more can reflect
The details difference of every image, is modified the deconvolution parameter in convolutional neural networks, obtains revised target convolution mind
Through network.
Modified method is referred to the prior art, for the quantification and precision for guaranteeing makeover process, it is preferable that repair
Just can be by constructing loss function, loss function is smaller, and the fitting of network is better.It then specifically can be according to similarity to convolution
Parameter be modified including：
The loss function of convolutional neural networks is constructed according to similarity；
It is modified according to deconvolution parameter of the loss function to convolutional neural networks.
Being modified according to loss function can be can refer to it is not limited here using the method for common iterated revision
The prior art, since common correcting iteration method is than relatively time-consuming, to improve parameters revision efficiency, it is preferable that can be according to damage
Function is lost to be modified based on deconvolution parameter of the stochastic gradient descent method to convolutional neural networks.SGD (stochastic gradient descent)
Can optimization when accurate mathematical model can not be established, the method for iterative approach true value, constantly reduction mould
Type output error carries out near-optimal to single sample, and fast convergence rate, calculation amount is small, can greatly promote whole efficiency.
Step s200：Obtain topography.
Topography refers to a part in image only including whole things, such as the nasal portion in whole facial image
Image is topography.Without limitation to the acquisition methods of topography at this, topography can be directly inputted, it can also be with
Retain parts of images as needed after receiving general image, wherein the image pixel and resolution ratio of input need as far as possible
Height, in order to avoid influence feature extraction effect.
Wherein, optionally, obtaining topography can specifically include：
Receive general image；
Receive the local message for carrying out feature extraction；
Image is cut according to local message, obtains topography.
Step s300：The characteristic information of topography is extracted according to revised deconvolution parameter based on convolutional neural networks,
Obtain feature vector.
Topography is input in convolutional neural networks, convolutional neural networks according to deconvolution parameter accurate after amendment into
Row Automatic Feature Extraction, the corresponding feature vector of output topography.
It should be noted that being repaired in the step s100 of the offer of the present embodiment to the deconvolution parameter of convolutional neural networks
Positive process and it is non-required carrying out completing in local image characteristics extraction process every time, when deconvolution parameter amendment reaches default standard
When true rate, step s100 can not be executed, and only need to execute step according to revised accurate deconvolution parameter to the extraction of feature
S200 and s300.
Based on above-mentioned introduction, local image characteristics provided by the present application, which are extracted, extracts topography by convolutional neural networks
Characteristic information, convolutional neural networks can be found that and portray complicated structure feature inside topography, so can be larger
Improve the performance of feature extraction；In addition, being obtained by that will include different classes of topography's information input to convolutional neural networks
To after feature vector, the corresponding feature vector of each topography's information is found by way of calculating characteristic central point mutually similar
In not and the similarity degree of different classes of characteristics of image, by similarity degree assess learning network in similar image and
Differentiation degree between inhomogeneity image, migrates inhomogeneity image characteristic extracting method, mentions to similar local image characteristics
It takes and distinguishes, deconvolution parameter is constantly corrected, obtain the deconvolution parameter that can more precisely portray details distinguishing characteristics, repair
Positive process only needs to calculate similarity degree, and calculation amount is smaller, and makeover process is rapid, the revised convolutional Neural net of deconvolution parameter
Network can more accurately excavate corresponding class another characteristic, and image recognition rate is high, and robustness is high.
For ease of understanding, entire deconvolution parameter is repaired so that the convolutional neural networks of foundation are seven layers of convolutional layer as an example herein
Positive process is introduced, and other way can refer to the introduction of the present embodiment.
It can specifically include：
Step s1.1：K classification is randomly selected from the local image set of training, n images are randomly selected from each class；
Step s1.2：The image obtained in the step s1.1 is separately input to the Layer1 convolution in convolutional neural networks
Layer, Layer1 convolutional layer include 32 3 × 3 convolution kernels, and step-length is 1 pixel, carry out convolution operation simultaneously to the image of input
Make batch standardization (batch normalization), the use of ReLU is activation primitive, the Layer1 convolution of output 32 × 32 × 32
Layer characteristic image；
Step s1.3：Layer2 convolutional layer, Layer2 will be input to from convolutional layer characteristic image obtained in step s1.2
Convolutional layer includes 32 3 × 3 convolution kernels, and step-length is 1 pixel, carries out convolution operation to the image of input and makees batch
Normalization is activation primitive, the Layer2 convolutional layer characteristic image of output 32 × 32 × 32 using ReLU；
Step s1.4：Layer3 convolutional layer, Layer3 will be input to from convolutional layer characteristic image obtained in step s1.3
Convolutional layer includes 64 3 × 3 convolution kernels, and step-length is 2 pixels, carries out convolution operation to the image of input and makees batch
Normalization is activation primitive, the Layer3 convolutional layer characteristic image of output 64 × 32 × 32 using ReLU；
Step s1.5：Layer4 convolutional layer, Layer4 will be input to from convolutional layer characteristic image obtained in step s1.4
Convolutional layer includes 64 3 × 3 convolution kernels, and step-length is 1 pixel, carries out convolution operation to the image of input and makees batch
Normalization is activation primitive, the Layer4 convolutional layer characteristic image of output 64 × 32 × 32 using ReLU；
Step s1.6：Layer5 convolutional layer, Layer5 will be input to from convolutional layer characteristic image obtained in step s1.5
Convolutional layer includes 128 3 × 3 convolution kernels, and step-length is 2 pixels, carries out convolution operation to the image of input and makees batch
Normalization is activation primitive, the Layer5 convolutional layer characteristic image of output 128 × 32 × 32 using ReLU；
Step s1.7：Layer6 convolutional layer, Layer6 will be input to from convolutional layer characteristic image obtained in step s1.6
Convolutional layer includes 128 3 × 3 convolution kernels, and step-length is 1 pixel, carries out convolution operation to the image of input and makees batch
Normalization is activation primitive using ReLU, exports 128 × 32 × 32 Layer5 convolutional layer characteristic image, and is arranged
Random inactivation factor is 0.25；
Step s1.8：Layer7 convolutional layer, Layer7 will be input to from convolutional layer characteristic image obtained in step s1.7
Convolutional layer includes 128 8 × 8 convolution kernels, carries out convolution operation to the image of input and makees batch normalization,
The character representation vector of 128 dimension of output；
Step s2.1：If the k obtained in the step s1.8 is classified as A1,A2,A3,...,Ak, wherein each sorting group i
IncludeEqual n character representation vector.
Step s2.2：The central point of each sorting group in step s2.1 is calculatedCalculation formula is such as
Under
Step s2.3：Calculate all the points in every groupWith central point in groupEuclidean distanceCalculation formula is as follows
Step s2.4：Calculate central point in every groupWith central point in remaining k-1 groupEuclidean distance
Dij, calculation formula is as follows
Step s2.5：For the central point of any one group iFinding in one group of data obtained in the step s2.4 makes
Obtain DijThe smallest group of j, i.e.,
And set the central point character representation vector of the j group asIf their Euclidean distance is
Step s3：It is obtained according to step s2.3The D that step s2.4 is obtainedijIt is obtained with step s2.5Constructing loss function is
Step s4：Optimize convolutional Neural net using the method for stochastic gradient descent according to the loss function that step s3 is obtained
Network parameter, repeats the above steps, until the loss function of step s3 no longer becomes smaller or stablizes, obtains including revised convolution
The convolutional neural networks of parameter.
The amendment of parameter is carried out through the above steps, and makeover process efficiency is very high, corrects to obtain by the inclusion of above-mentioned steps
The neural network of deconvolution parameter carry out feature extraction, extraction effect is preferable.
Local image characteristics extraction element provided by the present application is introduced below, referring to FIG. 4, Fig. 4 is the application
The structural block diagram for the local image characteristics extraction element that embodiment provides；The apparatus may include：
Amending unit 100 obtains revised convolution ginseng for being modified to the deconvolution parameter in convolutional neural networks
Number；
Acquiring unit 200, for obtaining topography；
Feature extraction unit 300, for being mentioned based on convolutional neural networks according to the revised deconvolution parameter of amending unit 100
The characteristic information for taking topography, obtains feature vector；
Wherein, the structural block diagram of amending unit 100 is as shown in figure 5, mainly include：
Image inputs subelement 110, includes that different classes of topography's information input is refreshing to convolution for what be will acquire
Through network, the corresponding feature vector of each topography's information is obtained；
Central point computation subunit 120, for calculating the corresponding characteristic central point of topography's information of all categories；
Similarity calculation subelement 130, for calculating part in of all categories and same category according to characteristic central point
The similarity of image；
Parameters revision subelement 140, for being modified according to similarity to deconvolution parameter.
It should be noted that each unit in local image characteristics extraction element in the application specific embodiment,
Its course of work please refers to the corresponding specific embodiment of Fig. 1, and details are not described herein.
Optionally, 130 structural block diagram of similarity calculation subelement is as shown in fig. 6, can specifically include：
First computation subunit 131, for calculating the Europe of each feature vector and corresponding characteristic central point in same category
Distance is obtained in several；
Subelement 132 is counted, the fluctuation situation of Euclidean distance is corresponded to for counting each feature vector, as corresponding class
The similarity of Bie Nei topography；
Second computation subunit 133, for calculating the Euclidean distance between variant classification between characteristic central point, as
The similarity of of all categories topography.
Wherein, statistics subelement 132 specifically can be used for calculating each feature vector and correspond to variance between Euclidean distance.
Optionally, the structural block diagram of parameters revision subelement 140 is as shown in fig. 7, mainly include：
Loss function constructs subelement 141, for constructing the loss function of convolutional neural networks according to similarity；
Loss function revise subelemen 142, for being repaired according to deconvolution parameter of the loss function to convolutional neural networks
Just.
Wherein, loss function revise subelemen 142 specifically can be used for being based on stochastic gradient descent side according to loss function
Method is modified the deconvolution parameter of convolutional neural networks.
Optionally, the structural block diagram of acquiring unit 200 is as shown in figure 8, can specifically include：
Receiving subelement 210, for receiving general image；
Subelement 220 is extracted, for receiving the local message for carrying out feature extraction；
It cuts subelement 230 and obtains topography for cutting according to local message to image.
Readable storage medium storing program for executing provided by the embodiments of the present application is introduced below, readable storage medium storing program for executing described below with
Above-described local image characteristics extracting method can correspond to each other reference.
A kind of local image characteristics extraction system is also disclosed in the application, mainly includes：Memory and processor.
Wherein, memory is for storing computer program；
The step of processor is for realizing above-mentioned local image characteristics extracting method when executing computer program.
A kind of computer readable storage medium is also disclosed in the application, program is stored thereon with, when program is executed by processor
The step of realizing local image characteristics extracting method.
It is apparent to those skilled in the art that for convenience and simplicity of description, the device of foregoing description,
The specific work process of equipment, storage medium and unit, can refer to corresponding processes in the foregoing method embodiment, herein no longer
It repeats.
In several embodiments provided herein, it should be understood that disclosed device, system, storage medium and
Method may be implemented in other ways.For example, the apparatus embodiments described above are merely exemplary, for example, single
Member division, only a kind of logical function partition, there may be another division manner in actual implementation, for example, multiple units or
Component can be combined or can be integrated into another system, or some features can be ignored or not executed.Another point is shown
The mutual coupling, direct-coupling or communication connection shown or discussed can be through some interfaces, between device or unit
Coupling or communication connection are connect, can be electrical property, mechanical or other forms.
Unit may or may not be physically separated as illustrated by the separation member, shown as a unit
Component may or may not be physical unit, it can and it is in one place, or may be distributed over multiple networks
On unit.It can some or all of the units may be selected to achieve the purpose of the solution of this embodiment according to the actual needs.
It, can also be in addition, each functional unit in each embodiment of the application can integrate in one processing unit
It is that each unit physically exists alone, can also be integrated in one unit with two or more units.Above-mentioned integrated list
Member both can take the form of hardware realization, can also realize in the form of software functional units.
It, can if integrated unit is realized in the form of SFU software functional unit and when sold or used as an independent product
To be stored in a mobile terminal.Based on this understanding, the technical solution of the application is substantially in other words to the prior art
The all or part of the part to contribute or the technical solution can be embodied in the form of software products, which deposits
It stores up in one storage medium, including some instructions are used so that a mobile terminal (can be mobile phone or tablet computer
Deng) execute each embodiment method of the application all or part of the steps.And storage medium above-mentioned includes：USB flash disk, movement are hard
Disk, read-only memory (Read-Only Memory, ROM), random access memory (Random Access Memory, RAM),
The various media that can store program code such as magnetic or disk.
Each embodiment is described in a progressive manner in specification, the highlights of each of the examples are with other realities
The difference of example is applied, the same or similar parts in each embodiment may refer to each other.For device disclosed in embodiment
Speech, since it is corresponded to the methods disclosed in the examples, so being described relatively simple, related place is referring to method part illustration
?.
Professional further appreciates that, unit described in conjunction with the examples disclosed in the embodiments of the present disclosure
And algorithm steps, it can be realized with the combination of electronic hardware, terminal or the two, in order to clearly demonstrate hardware and software
Interchangeability generally describes each exemplary composition and step according to function in the above description.These functions are studied carefully
Unexpectedly it is implemented in hardware or software, the specific application and design constraint depending on technical solution.Professional technique people
Member can use different methods to achieve the described function each specific application, but this realization is it is not considered that super
Scope of the present application out.
The step of method described in conjunction with the examples disclosed in this document or algorithm, can directly be held with hardware, processor
The combination of capable software module or the two is implemented.Software module can be placed in random access memory (RAM), memory, read-only deposit
Reservoir (ROM), electrically programmable ROM, electrically erasable ROM, register, hard disk, moveable magnetic disc, CD-ROM or technology
In any other form of storage medium well known in field.
Above to local image characteristics extracting method, device, system and computer-readable storage medium provided herein
Matter is described in detail.Specific examples are used herein to illustrate the principle and implementation manner of the present application, above
The explanation of embodiment is merely used to help understand the present processes and its core concept.It should be pointed out that for the art
Those of ordinary skill for, under the premise of not departing from the application principle, can also to the application carry out it is several improvement and repair
Decorations, these improvement and modification are also fallen into the protection scope of the claim of this application.
Claims (10)
1. a kind of local image characteristics extracting method, which is characterized in that including：
Deconvolution parameter in convolutional neural networks is modified, revised deconvolution parameter is obtained；
Obtain topography；
The characteristic information for being extracted the topography according to the revised deconvolution parameter based on the convolutional neural networks, is obtained
To feature vector；
Wherein, the modification method of the deconvolution parameter includes：
What be will acquire includes different classes of topography's information input to convolutional neural networks, obtains each topography's information pair
The feature vector answered；
Calculate the corresponding characteristic central point of topography's information of all categories；
The similarity of topography in of all categories and same category is calculated according to the characteristic central point；
The deconvolution parameter is modified according to the similarity.
2. local image characteristics extracting method as described in claim 1, which is characterized in that described according to the characteristic central point
The similarity for calculating topography in of all categories and same category includes：
Calculate the Euclidean distance of each feature vector and corresponding characteristic central point in same category；
The fluctuation situation that each feature vector corresponds to Euclidean distance is counted, the similarity as topography in corresponding classification；
Calculate the Euclidean distance between variant classification between characteristic central point, the similarity as of all categories topography.
3. local image characteristics extracting method as claimed in claim 2, which is characterized in that each feature vector of statistics is corresponding
The fluctuation situation of Euclidean distance includes：
It calculates each feature vector and corresponds to variance between Euclidean distance.
4. local image characteristics extracting method as described in claim 1, which is characterized in that it is described according to the similarity to institute
State deconvolution parameter be modified including：
The loss function of the convolutional neural networks is constructed according to the similarity；
It is modified according to the deconvolution parameter of the loss function to the convolutional neural networks.
5. local image characteristics extracting method as claimed in claim 4, which is characterized in that described according to the loss function pair
The deconvolution parameter of the convolutional neural networks be modified including：
It is carried out according to the loss function based on the deconvolution parameter of the stochastic gradient descent method to the convolutional neural networks
Amendment.
6. local image characteristics extracting method as described in claim 1, which is characterized in that the acquisition topography includes：
Receive general image；
Receive the local message for carrying out feature extraction；
Described image is cut according to the local message, obtains topography.
7. a kind of local image characteristics extraction element, which is characterized in that including：
Amending unit obtains revised deconvolution parameter for being modified to the deconvolution parameter in convolutional neural networks；
Acquiring unit, for obtaining topography；
Feature extraction unit extracts the topography according to the revised deconvolution parameter based on the convolutional neural networks
Characteristic information, obtain feature vector；
Wherein, the amending unit includes：
Image inputs subelement, includes different classes of topography's information input to convolutional neural networks for what be will acquire,
Obtain the corresponding feature vector of each topography's information；
Central point computation subunit, for calculating the corresponding characteristic central point of topography's information of all categories；
Similarity calculation subelement, for calculating topography in of all categories and same category according to the characteristic central point
Similarity；
Parameters revision subelement, for being modified according to the similarity to the deconvolution parameter.
8. local image characteristics extraction element as claimed in claim 7, which is characterized in that the similarity calculation subelement packet
It includes：
First computation subunit, for calculate in same category each feature vector and the Euclid of corresponding characteristic central point away from
From；
Subelement is counted, the fluctuation situation of Euclidean distance is corresponded to for counting each feature vector, as office in corresponding classification
The similarity of portion's image；
Second computation subunit, for calculating the Euclidean distance between variant classification between characteristic central point, as of all categories
Between topography similarity.
9. a kind of local image characteristics extraction system, which is characterized in that including：
Memory, for storing computer program；
Processor realizes that the topography as described in claim 1 to 6 any one is special when for executing the computer program
The step of levying extracting method.
10. a kind of computer readable storage medium, which is characterized in that it is stored with program on the computer readable storage medium,
It is realized when described program is executed by processor as described in any one of claim 1 to 6 the step of local image characteristics extracting method.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810870557.9A CN108921181B (en) | 2018-08-02 | 2018-08-02 | Local image feature extraction method, device and system and readable storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810870557.9A CN108921181B (en) | 2018-08-02 | 2018-08-02 | Local image feature extraction method, device and system and readable storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108921181A true CN108921181A (en) | 2018-11-30 |
| CN108921181B CN108921181B (en) | 2022-05-10 |
Family
ID=64394361
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810870557.9A Active CN108921181B (en) | 2018-08-02 | 2018-08-02 | Local image feature extraction method, device and system and readable storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108921181B (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109567600A (en) * | 2018-12-05 | 2019-04-05 | 江西书源科技有限公司 | The accessory automatic identifying method of household water-purifying machine |
| CN109829350A (en) * | 2019-02-20 | 2019-05-31 | 北京意锐新创科技有限公司 | The two dimensional code method of payment and device of compatible for wired and wireless two-channel access public network |
| CN109978090A (en) * | 2019-03-18 | 2019-07-05 | 北京意锐新创科技有限公司 | The method of payment and device combined based on two dimensional code and private mark |
| CN109977717A (en) * | 2019-03-18 | 2019-07-05 | 北京意锐新创科技有限公司 | Bar code method of payment and device based on light and shade code |
| WO2020133510A1 (en) * | 2018-12-29 | 2020-07-02 | 深圳迈瑞生物医疗电子股份有限公司 | Ultrasonic imaging method and device |
| CN113505740B (en) * | 2021-07-27 | 2023-10-10 | 北京工商大学 | Face recognition method based on transfer learning and convolutional neural network |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101369316A (en) * | 2008-07-09 | 2009-02-18 | 东华大学 | Image characteristics extraction method based on global and local structure amalgamation |
| CN101853398A (en) * | 2010-05-11 | 2010-10-06 | 浙江大学 | Chinese paper cutting identification method based on space constraint characteristic selection and combination thereof |
| CN106503743A (en) * | 2016-10-31 | 2017-03-15 | 天津大学 | A kind of quantity is more and the point self-adapted clustering method of the high image local feature of dimension |
| CN107403166A (en) * | 2017-08-02 | 2017-11-28 | 广东工业大学 | A kind of method and apparatus for extracting facial image pore feature |
-
2018
- 2018-08-02 CN CN201810870557.9A patent/CN108921181B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101369316A (en) * | 2008-07-09 | 2009-02-18 | 东华大学 | Image characteristics extraction method based on global and local structure amalgamation |
| CN101853398A (en) * | 2010-05-11 | 2010-10-06 | 浙江大学 | Chinese paper cutting identification method based on space constraint characteristic selection and combination thereof |
| CN106503743A (en) * | 2016-10-31 | 2017-03-15 | 天津大学 | A kind of quantity is more and the point self-adapted clustering method of the high image local feature of dimension |
| CN107403166A (en) * | 2017-08-02 | 2017-11-28 | 广东工业大学 | A kind of method and apparatus for extracting facial image pore feature |
Non-Patent Citations (3)
| Title |
|---|
| ANASTASIYA MISHCHUK ET AL: "Working hard to know your neighbor’s margins:Local descriptor learning loss", 《ARXIV.ORG》 * |
| XINWEI HE ET AL: "Triplet-Center Loss for Multi-View 3D Object Retrieval", 《ARXIV.ORG》 * |
| 孙祝岭 著: "《时间序列与多元统计分析》", 30 September 2016, 上海交通大学出版社 * |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109567600A (en) * | 2018-12-05 | 2019-04-05 | 江西书源科技有限公司 | The accessory automatic identifying method of household water-purifying machine |
| WO2020133510A1 (en) * | 2018-12-29 | 2020-07-02 | 深圳迈瑞生物医疗电子股份有限公司 | Ultrasonic imaging method and device |
| CN112672691A (en) * | 2018-12-29 | 2021-04-16 | 深圳迈瑞生物医疗电子股份有限公司 | Ultrasonic imaging method and equipment |
| CN112672691B (en) * | 2018-12-29 | 2024-03-29 | 深圳迈瑞生物医疗电子股份有限公司 | Ultrasonic imaging method and equipment |
| CN109829350A (en) * | 2019-02-20 | 2019-05-31 | 北京意锐新创科技有限公司 | The two dimensional code method of payment and device of compatible for wired and wireless two-channel access public network |
| CN109978090A (en) * | 2019-03-18 | 2019-07-05 | 北京意锐新创科技有限公司 | The method of payment and device combined based on two dimensional code and private mark |
| CN109977717A (en) * | 2019-03-18 | 2019-07-05 | 北京意锐新创科技有限公司 | Bar code method of payment and device based on light and shade code |
| CN113505740B (en) * | 2021-07-27 | 2023-10-10 | 北京工商大学 | Face recognition method based on transfer learning and convolutional neural network |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108921181B (en) | 2022-05-10 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108921181A (en) | A kind of local image characteristics extracting method, device, system and readable storage medium storing program for executing | |
| US11222211B2 (en) | Method and apparatus for segmenting video object, electronic device, and storage medium | |
| CN110826519A (en) | Face occlusion detection method and device, computer equipment and storage medium | |
| CN111160229B (en) | SSD network-based video target detection method and device | |
| CN109740679B (en) | Target identification method based on convolutional neural network and naive Bayes | |
| CN112541434B (en) | Face recognition method based on central point tracking model | |
| CN112465801B (en) | Instance segmentation method for extracting mask features in scale division mode | |
| EP3012779A1 (en) | Method and apparatus for extracting feature correspondences from multiple images | |
| CN112488923A (en) | Image super-resolution reconstruction method and device, storage medium and electronic equipment | |
| CN107679539B (en) | Single convolution neural network local information and global information integration method based on local perception field | |
| CN112132279A (en) | Convolutional neural network model compression method, device, equipment and storage medium | |
| CN111415373A (en) | Target tracking and segmenting method, system and medium based on twin convolutional network | |
| CN114419349A (en) | Image matching method and device | |
| CN111339884A (en) | Image recognition method and related equipment and device | |
| CN110135428B (en) | Image segmentation processing method and device | |
| CN109857895B (en) | Stereo vision retrieval method and system based on multi-loop view convolutional neural network | |
| CN113128518B (en) | Sift mismatch detection method based on twin convolution network and feature mixing | |
| CN111881947B (en) | Feature point set pair screening method, target matching method, device and electronic equipment | |
| CN116342466A (en) | Image matting method and related device | |
| CN113610856A (en) | Method and device for training image segmentation model and image segmentation | |
| CN111862159A (en) | Improved target tracking and segmentation method, system and medium for twin convolutional network | |
| CN111126177A (en) | People counting method and device | |
| Li et al. | MA-NET: Multi-scale attention-aware network for optical flow estimation | |
| CN111340084B (en) | Data processing method and device, processor, electronic equipment and storage medium | |
| US11989927B2 (en) | Apparatus and method for detecting keypoint based on deep learning using information change across receptive fields |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |