CN107992800A - A kind of fingerprint image quality determination methods based on SVM and random forest - Google Patents
A kind of fingerprint image quality determination methods based on SVM and random forest Download PDFInfo
- Publication number
- CN107992800A CN107992800A CN201711103387.3A CN201711103387A CN107992800A CN 107992800 A CN107992800 A CN 107992800A CN 201711103387 A CN201711103387 A CN 201711103387A CN 107992800 A CN107992800 A CN 107992800A
- Authority
- CN
- China
- Prior art keywords
- random forest
- sample
- svm
- feature
- training
- 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
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/12—Fingerprints or palmprints
- G06V40/1365—Matching; Classification
- G06V40/1376—Matching features related to ridge properties or fingerprint texture
-
- 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/74—Image or video pattern matching; Proximity measures in feature spaces
- G06V10/75—Organisation of the matching processes, e.g. simultaneous or sequential comparisons of image or video features; Coarse-fine approaches, e.g. multi-scale approaches; using context analysis; Selection of dictionaries
- G06V10/758—Involving statistics of pixels or of feature values, e.g. histogram matching
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Multimedia (AREA)
- Databases & Information Systems (AREA)
- Computing Systems (AREA)
- General Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Human Computer Interaction (AREA)
- Collating Specific Patterns (AREA)
Abstract
A kind of fingerprint image quality determination methods based on SVM and random forest, step include：Select the relatively good image of quality and be used as training positive sample, select second-rate image and be used as training negative sample；Feature extraction is carried out with negative sample with positive sample and training to training, these features are formed into feature vector and are normalized；Feature vector after normalization is input to LIBSVM to be trained to obtain SVM models；Feature vector after normalization is input to Random Forest model to be trained to obtain Random Forest model；Treat forecast sample and carry out feature extraction and feature vector normalized；The feature vector of normalized sample to be predicted is substituted into SVM models respectively and Random Forest model obtains corresponding mass fraction；The mass fraction that the mass fraction that SVM models obtain is obtained with Random Forest model is averaged, and as final prediction result.
Description
Technical field
The invention belongs to fingerprint field, is related to a kind of fingerprint image quality determination methods based on SVM and random forest.
Background technology
With the development of biological identification technology, fingerprint identification technology is because of its uniqueness, permanent and stability, extensively
Ground is applied to the various aspects that we live, and becomes a part indispensable during we live.But existing finger
Line identification module, due to therefrom, its recognition speed has much room for improvement with performance.And in registering fingerprint, feature extraction
Be to influence one of an important factor for whole fingerprint identification module is horizontal on the determined level of fingerprint image quality when comparing.Can
See, a kind of proposition for fast and effeciently judging fingerprint image quality method, speed and property to improving whole fingerprint identification module
There can be very big practical significance.
Under such a main trend, there is an urgent need for a kind of fingerprint image quality determination methods, quickly reject poor quality
Image, improves the recognition speed and performance of whole module.One preferable fingerprint identification module should require the multiple typing of user to refer to
Line, not registers for the fingerprint image of poor quality, only registers the preferable fingerprint image of quality.During feature extraction, only to that
A little preferable fingerprint image extraction features of quality, sincere and raising recognition speed is refused to be effectively reduced.When comparing, for matter
The fingerprint image of difference is measured, is not involved in comparing and generating template, accuracy of system identification can be so effectively reduced and improve recognition speed.
Scheme, which is described as follows, to be judged to the existing fingerprint image quality based on principal component analysis (PCA)：
1) to the fingerprint image piecemeal of input, multiple images block is obtained；
2) each pixel in the block to each image, obtains its neighborhood territory pixel information, blocking sample matrix；
3) projection matrix and feature value vector of each block sample matrix are obtained using principal component analytical method (PCA)；
4) feature for image block Quality estimation is calculated from projection matrix and feature value vector：Circle distribution characteristics and master
Direction character value residual error feature；
5) circle distribution characteristics is multiplied with principal direction characteristic value residual error feature, obtains the localized mass quality of image block；
6) the localized mass quality based on image block, with reference to image block Harris intensity, obtains the overall situation of input fingerprint image
Quality.
In conclusion the program needs piecemeal to judge the quality of fingerprint image, take larger, be not suitable for those to rate request
High application field.In addition, the Quality estimation accuracy rate of the program also has much room for improvement, the property of whole fingerprint identification module is reduced
Energy.Therefore, there is an urgent need for propose a kind of new fingerprint image quality determination methods.
Existing patent application CN104268529A discloses the determination methods and device of a kind of fingerprint image quality, described to sentence
Disconnected method includes step：Obtain fingerprint image sample；Learnt using SVM classifier according to the fingerprint image sample, obtained
Obtain optimal classification surface；Fingerprint image to be judged is obtained, and calculates the HOG features of the fingerprint image；According to the HOG features
The quality of the fingerprint image is judged with optimal classification surface.This application employs single machine learning method and combines single
Fingerprint characteristic, robustness is not high enough, and judging nicety rate also has much room for improvement.
The content of the invention
The present invention provides a kind of performance and speed for being greatly enhanced whole fingerprint identification module, judging result has more Shandong
The fingerprint image quality determination methods based on SVM and random forest of rod, accuracy rate higher.
The technical solution adopted by the present invention is：
A kind of fingerprint image quality determination methods based on SVM and random forest, it is characterised in that：Including sample training and
Sample predictions, sample training step include：
Select the relatively good image of quality and be used as training positive sample, select second-rate image and used as training and born
Sample；
Variance, ridge paddy contrast, orientation consistency, Gabor characteristic are carried out with negative sample with positive sample and training to training
These features are formed feature vector and are normalized by the extraction of value, HOG features and LBP features；
Feature vector after normalization is input to LIBSVM to be trained to obtain SVM models；
Feature vector after normalization is input to Random Forest model to be trained to obtain Random Forest model；
Sample predictions step includes：
Treat forecast sample and carry out feature extraction and feature vector normalized；
The feature vector of normalized sample to be predicted is substituted into SVM models respectively and Random Forest model obtains accordingly
Mass fraction；
The mass fraction that the mass fraction that SVM models obtain is obtained with Random Forest model is averaged, and as
Final prediction result.
Further, the preferable image of relative mass with the different finger collections that positive sample is different sensors is trained.
Further, train with negative sample be different sensors gather have finger figure without finger-image and poor quality
Picture.
Further, the feature extracted in sample predictions step with extracted in sample training step be characterized in it is consistent.
Further, normalization processing method and normalization processing method in sample training step are one in sample predictions step
Cause.
Further, feature extracting method is as follows in sample predictions step and sample training step：
A) average of view picture fingerprint image is first calculated, by the square cumulative of the difference of the gray value of each pixel and average
Get up, then divided by pixel number, obtain the variance of the image；
B) average of entire image is first calculated, then counts and is more than the pixel number of average in entire image and less than equal
The two, is finally divided by by the pixel number of value, obtains ridge paddy to comparing；
C) the gradient vector covariance matrix of entire image all pixels point is calculated first, then calculates the covariance matrix
Characteristic value a1 and a2, according to the characteristic value a1 and a2 being calculated, orientation consistency (coh) is represented by：
Coh=((a1-a2) * (a1-a2))/((a1+a2) * (a1+a2))；
D) extracting method of Gabor characteristic value is multiple for the mask convolution of image and different scale and direction is obtained
Gabor characteristic value, continues to calculate the standard deviation of above-mentioned Gabor characteristic value, which is that composition characteristic vector is required
Feature；
E) extracting method of HOG features is calculates gradient magnitude and the direction at each pixel, and by equidirectional pixel
The gradient magnitude of point adds up；The corresponding direction of greatest gradient amplitude and its opposite direction are found out, and the two directions are corresponded to
Gradient magnitude as the feature finally extracted.
F) by each pixel in image compared with 8 pixels of surrounding, if surrounding pixel values are more than center pixel value,
Then the position of the pixel is marked as 1, is otherwise 0, and 8 points in such 3*3 neighborhoods can produce 8 binary systems through comparing
Number, that is, obtain the decimal system LBP values of window center pixel；The frequency that each decimal system LBP values occur is counted, generates histogram
And be normalized, the histogram after the normalization is the LBP features of required extraction.
Further, SVM models include training used by kernel function, penalty coefficient, support vector machines number, support to
The bias term of amount and discriminant function.
Further, Random Forest model includes the number of the decision tree of composition random forest, the node split category of decision tree
Property and node decision function, it is that it is exported to be completed the result is that passing through most ballot comparative analyses.
Beneficial effects of the present invention：
1) Quality estimation is carried out to view picture fingerprint image, rather than piecemeal is judged, substantially increases fingerprint image image quality
The speed judged is measured, so as to improve the recognition speed of whole fingerprint identification module.
2) Quality estimation is carried out compared to using single machine learning method, the present invention uses two kinds of machine learning methods
The mode being combined is judged that judging nicety rate greatly improves, not only reduce whole fingerprint identification module refuse it is sincere
Reduce accuracy of system identification.
3) the technology used in the present invention judges that speed is fast and low to space power consumption requirements, right particularly suitable for those
The occasion that rate request height and condition are limited.
Brief description of the drawings
Fig. 1 is the sample training flow chart of the present invention.
Fig. 2 is the sample predictions flow chart of the present invention.
Fig. 3 is the Quality estimation result schematic diagram of the fingerprint image of the different quality of the present invention.
Embodiment
The present invention is further described with reference to specific embodiment, but does not limit the invention to these tools
Body embodiment.One skilled in the art would recognize that present invention encompasses may include in Claims scope
All alternatives, improvement project and equivalents.
Respective explanations are carried out to technical term of the present invention, including：
SVM (support vector machines)：The original sorting algorithm with obvious intuitive geometry meaning, with higher accurate
Rate.Its thought is directly perceived, but details complex, and content is related to convextiry analysis algorithm, the field of the profundity such as kernel function and neutral net.
SVM is a kind of algorithm for allowing applied mathematics to be really applied.
Random forest (Random Forest)：It is composed of many decision tree classifiers, single decision tree classifier
Formed with random device, thus be referred to as " random forest ".It more has robustness to mistake and outlier, in big data feelings
Speed is fast under condition, and performance is good.
Principal component analysis (PCA)：By carrying out linear combination, the index optimized to original variable.Original multiple
The dimensionality reduction that calculates of index is a small amount of several calculating by optimizing index.Its basic thought is：Try have one by originally numerous
Determine the index of correlation, reconfigure as one group of new mutually independent overall target, and replace original index.This method
It is one kind of machine learning method.
Harris：A kind of critical point detection algorithm, is a very important algorithm in image detection identification, to object appearance
Gesture change robustness is good, insensitive to rotating, and can detect the angle point of object well.
Gabor：In image processing field, Gabor filter is a linear filter for being used for edge detection.Gabor
The frequency of wave filter and direction represent the expression close to human visual system for frequency and direction, therefore are usually used in texture representation
And description.In spatial domain, the Gabor filter of one 2 dimension is the product of sinusoidal a plane wave and gaussian kernel function.Practical application
Middle Gabor filter can extract correlated characteristic on the different scale and different directions of frequency domain.
HOG：That is histograms of oriented gradients feature, is that one kind is used for carrying out object inspection in computer vision and image procossing
The Feature Descriptor of survey.It is by calculating the gradient orientation histogram with statistical picture regional area come constitutive characteristic.
LBP：It is a kind of operator for being used for describing image local textural characteristics, it has rotational invariance and gray scale consistency
The features such as, it is the operator that image characteristics extraction is often used.
Refuse sincere (FRR):The different images of same finger collection are identified as the probability of different fingers.
Accuracy of system identification (FAR):The different images of different finger collections are identified as the probability of same finger.
Machine learning techniques based on SVM：Sample is chosen first, using the preferable fingerprint image of quality as positive sample, matter
Poor fingerprint image is measured as negative sample.Secondly, to each sample extraction feature, feature vector is generated, and returned
One changes.Using LIBSVM (LIBSVM be one of the exploitation design such as Taiwan Univ.'s woods intelligence benevolence (Lin Chih-Jen) professor it is simple,
It is easy to use and fast and effectively SVM pattern-recognitions and return software kit) feature vector after normalization is trained, obtain
To training pattern.This model includes kernel function, penalty coefficient, the number of support vector machines, supporting vector used by training
With the bias term of discriminant function etc..
Machine learning techniques based on random forest:The sample of the technology, which is chosen, feature generates and normalization is same is based on SVM
Machine learning techniques.Feature vector after normalization is put into random forest tool box to be trained, obtains training pattern.Should
Training pattern includes the number of decision tree, the node split attribute of decision tree and node decision function of composition random forest etc..
The output of random forest is divided target sample by n decision tree the result is that by majority ballot comparative analysis completion
Class decision-making, collects the output result of all decision trees, and the overall output of random forest is measured out by comparing votes
As a result.
Referring to Fig. 1 and Fig. 2, a kind of fingerprint image quality judgement side based on SVM and random forest proposed by the invention
Method can be divided into sample training and sample predictions two large divisions.
A kind of fingerprint image quality determination methods based on SVM and random forest are described as follows：
1) positive negative sample is selected：The preferable image of relative mass of the different finger collections of different sensors is selected as instruction
White silk positive sample, that selects different sensors collection has finger-image as training with negative sample without finger-image and poor quality
This.Here selecting for positive negative sample should be representative, covers more situations as far as possible, is otherwise easy to cause over-fitting and model
The problems such as generalization ability is poor.
2) sample characteristics extraction and normalization：Its variance, ridge paddy are calculated each image respectively first to comparison, direction one
Then these features are formed feature vector and are normalized by cause property, Gabor characteristic value, HOG features and LBP features.
The extracting method of all of above feature is specifically introduced as follows：
A) variance：The average of view picture fingerprint image is first calculated, by the flat of the difference of the gray value of each pixel and average
Side add up, then divided by pixel number, be the image variance.
B) ridge paddy contrast：The average of entire image is first calculated, then counts the pixel for being more than average in entire image
The two, is finally divided by as ridge paddy to comparing by number and the pixel number less than average.
C) orientation consistency：The gradient vector covariance matrix of entire image all pixels point is calculated first, is then calculated
The characteristic value a1 and a2 of the covariance matrix.According to the characteristic value a1 and a2 being calculated, orientation consistency (coh) can represent
For：
Coh=((a1-a2) * (a1-a2))/((a1+a2) * (a1+a2)).
D) Gabor characteristic value：The extracting method of this feature is to obtain image and the mask convolution in different scale and direction
Multiple Gabor characteristic values, continue to calculate the standard deviation of above-mentioned Gabor characteristic value, which is needed for composition characteristic vector
The feature wanted.
E) HOG features：Calculate the gradient magnitude at each pixel and direction, and by the gradient magnitude of equidirectional pixel
Add up；The corresponding direction of greatest gradient amplitude and its opposite direction are found out, and the corresponding gradient magnitude in the two directions is made
For the feature finally extracted.
F) LBP features：By each pixel is compared with 8 pixels of surrounding in image, if during surrounding pixel values are more than
Heart pixel value, then the position of the pixel be marked as 1, be otherwise 0.8 points so in 3*3 neighborhoods can produce 8 through comparing
Bit, that is, obtain the decimal system LBP values of window center pixel.The frequency that each decimal system LBP values occur is counted, it is raw
Into histogram and it is normalized.Histogram after the normalization is the LBP features of required extraction.
3) SVM is trained：By after normalization positive and negative sampling feature vectors input LIBSVM, adjusting parameter, by accuracy rate most
Training pattern when high is as optimal models.This model includes kernel function, penalty coefficient, support vector machines used by training
Number, the bias term of supporting vector and discriminant function etc..SVM models are described in detail below：
SVM types:For the SVM types that the present invention uses for C- support vector classifications, parameter C is penalty coefficient, and C is more big right
The punishment of mistake classification is bigger, and appropriate parameter C is most important to the accuracy rate for improving classification, and general acquiescence takes 1.In reality
The size of C can be constantly adjusted in training process, and selects the C for making training sample obtain highest accuracy rate, its general value is more than
1。
Kernel function：For RBF kernel functions, its expression formula is the kernel function that the present invention uses:
K (u, v)=exp (- gamma* | | u-v | | * | | u-v | |).
Gamma is the parameter of RBF kernel functions, its accuracy rate to category of model has considerable influence, and default value is characterized dimension
The inverse of degree.The size of gamma can be constantly adjusted in actual training process, and selects to make training sample acquirement highest accurate
The gamma of rate, its general value are more than the inverse of characteristic dimension.By RBF kernel functions, sample is projected into higher dimensional space, by line
The inseparable situation of property is changed into linear separability, substantially increases the accuracy rate of classification.
The number nsv of supporting vector：The number of general supporting vector should not be too large, and preferably not more than 10000, otherwise can
Cause predicted time long, it is difficult to practical application.
Supporting vector sv：The dimension of supporting vector is equal to characteristic dimension, has corresponding supporting vector per one kind.Such as 2 points
Supporting vector is divided into 2 parts by class, and supporting vector is divided into 3 parts by 3 classification, and so on.
Discriminant function bias term b：The introducing of the bias term, is conducive to obtain optimal separating plane.
4) random forest is trained：By the positive and negative sampling feature vectors input random forest tool box after normalization, adjustment ginseng
Number, obtains optimal training pattern.The training pattern includes the number of the decision tree of composition random forest, the node point of decision tree
Split attribute and node decision function etc..The output of random forest by majority ballot comparative analysis the result is that completed, i.e., target
Sample carries out categorised decision by n decision tree, the output result of all decision trees is collected, by comparing ballot quantity
Draw the overall output result of random forest.Random Forest model is described in detail below：
The number of decision tree:The reference record random forest is made of a how many decision tree.
The number of nodes of decision tree:Every decision tree of the reference record is made of how many node.
Node split attribute：The Split Attribute of each node of the reference record, the original that the present invention passes through information gain maximum
Then select optimal Split Attribute.
Node decision function：The optimal separation threshold value of the decision function is also come selection, its calculating according to information gain
Formula is as follows：
Gain (A)=Info (D)-InfoA(D)
Wherein Gain (A) is information gain, it have recorded the change of comentropy.
Voting results：The voting results of every decision tree are counted, it is final classification results to possess the most class of poll.
5) sample characteristics extraction and normalization to be predicted：The feature of fingerprint image to be predicted is extracted first, then will extraction
Feature composition feature vector and be normalized.Here the feature extracted must be consistent with the feature extracted during training, and returns
One change method also should be consistent with the method for normalizing used during training.
6) SVM is predicted：Feature vector after normalization is substituted into SVM models, obtains mass fraction.In order to applied to fingerprint
Identification module, the anticipation function are that C language is realized.
7) random forest is predicted：Feature vector after normalization is substituted into Random Forest model, obtains mass fraction.Together
SVM predicts that the anticipation function is also what C language was realized.
8) final prediction result：The mass fraction that the mass fraction that SVM models obtain is obtained with Random Forest model is asked
Average, and as final prediction result.
Quality estimation result example such as Fig. 3 institutes of the fingerprint image for the different quality that the present embodiment gathers different sensors
Show, example is judged from picture quality, the present invention without finger-image and has the equal energy of finger-image to different sensors collection
Obtain more rational fraction.
The present invention extracts feature such as Gabor characteristic used in training and prediction with different feature extracting methods from different perspectives
Value, HOG features and LBP features so that the feature vector of extraction has more distinction, can obtain more preferably training pattern；It is right
View picture fingerprint image extracts feature, rather than piecemeal extraction feature, while judging nicety rate is not reduced, can improve judgement speed
Degree；No longer judged using single machine learning method, but two kinds of machine learning methods are combined, judging result
With more robustness, accuracy rate higher.
The application scenario of the present invention：
1) fingerprint register：For the image of no finger-image and poor quality, not register in registration phase, can so carry
The performance of high whole fingerprint identification module.
2) Finger print characteristic abstract：In feature extraction phases, the image of those poor qualities can be picked out by Quality estimation,
Feature is not extracted to this parts of images, is advantageously reduced and is refused sincere and improve recognition speed.
3) compare：, will be without comparing and generating template, to reduce for the feature of the image zooming-out of those poor qualities
Accuracy of system identification and raising recognition speed.
In conclusion a kind of fingerprint image quality determination methods based on SVM and random forest proposed by the invention are fitted
Close and be applied in fingerprint identification module, the occasion that and condition high to rate request particularly suitable for those is limited.Therefore it is suitable
The occasion that employing fingerprint identification module is identified is closed, technology proposed by the invention is all suitable for, it is seen that the present invention has very wide
Application prospect.
Claims (8)
- A kind of 1. fingerprint image quality determination methods based on SVM and random forest, it is characterised in that：Including sample training and sample This prediction, sample training step include：Select the relatively good image of quality and be used as training positive sample, select second-rate image as the negative sample of training This；To training with positive sample and it is trained with negative sample carry out variance, ridge paddy contrast, orientation consistency, Gabor characteristic value, These features are formed feature vector and are normalized by the extraction of HOG features and LBP features；Feature vector after normalization is input to LIBSVM to be trained to obtain SVM models；Feature vector after normalization is input to Random Forest model to be trained to obtain Random Forest model；Sample predictions step includes：Treat forecast sample and carry out feature extraction and feature vector normalized；The feature vector of normalized sample to be predicted is substituted into SVM models respectively and Random Forest model obtains corresponding quality Fraction；The mass fraction that the mass fraction that SVM models obtain is obtained with Random Forest model is averaged, and as final Prediction result.
- 2. a kind of fingerprint image quality determination methods based on SVM and random forest according to claim 1, its feature exist In：Train the preferable image of relative mass with the different finger collections that positive sample is different sensors.
- 3. a kind of fingerprint image quality determination methods based on SVM and random forest according to claim 1, its feature exist In：Training with negative sample be different sensors gather have finger-image without finger-image and poor quality.
- 4. a kind of fingerprint image quality determination methods based on SVM and random forest according to claim 1, its feature exist In：The feature extracted in sample predictions step with extracted in sample training step be characterized in it is consistent.
- 5. a kind of fingerprint image quality determination methods based on SVM and random forest according to claim 1, its feature exist In：Normalization processing method and normalization processing method in sample training step are consistent in sample predictions step.
- 6. a kind of fingerprint image quality determination methods based on SVM and random forest according to one of Claims 1 to 5, It is characterized in that：Feature extracting method is as follows in sample predictions step and sample training step：A) average of view picture fingerprint image is first calculated, by square adding up for the difference of the gray value of each pixel and average Come, then divided by pixel number, obtain the variance of the image；B) average of entire image is first calculated, then counts and is more than the pixel number of average in entire image and less than average The two, is finally divided by by pixel number, obtains ridge paddy to comparing；C) the gradient vector covariance matrix of entire image all pixels point is calculated first, then calculates the spy of the covariance matrix Value indicative a1 and a2, according to the characteristic value a1 and a2 being calculated, orientation consistency (coh) is represented by：Coh=((a1-a2) * (a1-a2))/((a1+a2) * (a1+a2))；D) extracting method of Gabor characteristic value is that the mask convolution of image and different scale and direction is obtained multiple Gabor spies Value indicative, continues to calculate the standard deviation of above-mentioned Gabor characteristic value, which is the required feature of composition characteristic vector；E) extracting method of HOG features is calculates gradient magnitude and the direction at each pixel, and by equidirectional pixel Gradient magnitude adds up；Find out the corresponding direction of greatest gradient amplitude and its opposite direction, and by the corresponding ladder in the two directions Amplitude is spent as the feature finally extracted.F), should if surrounding pixel values are more than center pixel value by each pixel in image compared with 8 pixels of surrounding The position of pixel is marked as 1, is otherwise 0, and 8 points in such 3*3 neighborhoods can produce 8 bits through comparing, i.e., Obtain the decimal system LBP values of window center pixel；The frequency that each decimal system LBP values occur is counted, generation histogram is gone forward side by side Row normalization, the histogram after the normalization are the LBP features of required extraction.
- 7. a kind of fingerprint image quality determination methods based on SVM and random forest according to claim 6, its feature exist In：SVM models include kernel function, penalty coefficient, the number of support vector machines, supporting vector and discriminant function used by training Bias term.
- 8. a kind of fingerprint image quality determination methods based on SVM and random forest according to claim 6, its feature exist In：Random Forest model includes the number of decision tree, the node split attribute of decision tree and the node decision-making of composition random forest Function is that it is exported the result is that being completed by majority ballot comparative analysis.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711103387.3A CN107992800A (en) | 2017-11-10 | 2017-11-10 | A kind of fingerprint image quality determination methods based on SVM and random forest |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711103387.3A CN107992800A (en) | 2017-11-10 | 2017-11-10 | A kind of fingerprint image quality determination methods based on SVM and random forest |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107992800A true CN107992800A (en) | 2018-05-04 |
Family
ID=62030137
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201711103387.3A Pending CN107992800A (en) | 2017-11-10 | 2017-11-10 | A kind of fingerprint image quality determination methods based on SVM and random forest |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107992800A (en) |
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109166091A (en) * | 2018-06-29 | 2019-01-08 | 中国移动通信集团江苏有限公司 | Detection method, device, equipment and medium |
| CN109376628A (en) * | 2018-10-10 | 2019-02-22 | 杭州登虹科技有限公司 | A kind of picture quality detection method, device and storage medium |
| CN110318327A (en) * | 2019-06-10 | 2019-10-11 | 长安大学 | A kind of surface evenness prediction technique based on random forest |
| CN110472518A (en) * | 2019-07-24 | 2019-11-19 | 杭州晟元数据安全技术股份有限公司 | A kind of fingerprint image quality judgment method based on full convolutional network |
| CN111027629A (en) * | 2019-12-13 | 2020-04-17 | 国网山东省电力公司莱芜供电公司 | Power distribution network fault outage rate prediction method and system based on improved random forest |
| CN111127448A (en) * | 2019-12-26 | 2020-05-08 | 西南交通大学 | Model and method for detecting air spring fault based on isolated forest |
| CN112115917A (en) * | 2020-09-29 | 2020-12-22 | 深圳市汇顶科技股份有限公司 | Fingerprint identification method, fingerprint identification device, electronic equipment and storage medium |
| CN112287948A (en) * | 2020-10-30 | 2021-01-29 | 西安工程大学 | Sketch mark-based rapid edge detection method for extracting middle-level features |
| CN112926689A (en) * | 2021-03-31 | 2021-06-08 | 珠海格力电器股份有限公司 | Target positioning method and device, electronic equipment and storage medium |
| CN113361497A (en) * | 2021-08-09 | 2021-09-07 | 北京惠朗时代科技有限公司 | Intelligent tail box application method and device based on training sample fingerprint identification |
| CN115342900A (en) * | 2022-08-15 | 2022-11-15 | 东北石油大学 | Laser self-mixing interference micro-vibration measurement method and system based on random forest |
| CN112115917B (en) * | 2020-09-29 | 2024-05-28 | 深圳市汇顶科技股份有限公司 | Fingerprint identification method, fingerprint identification device, electronic equipment and storage medium |
Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20050063562A1 (en) * | 2003-08-07 | 2005-03-24 | Brunk Hugh L. | Conveying fingerprint minutiae with digital watermarks |
| EP1603077A1 (en) * | 2003-03-07 | 2005-12-07 | Nippon Telegraph and Telephone Corporation | Biological image correlation device and correlation method thereof |
| US20060120575A1 (en) * | 2004-12-04 | 2006-06-08 | Ahn Do S | Method and apparatus for classifying fingerprint image quality, and fingerprint image recognition system using the same |
| CN101303728A (en) * | 2008-07-01 | 2008-11-12 | 山东大学 | Method for identifying fingerprint facing image quality |
| CN103324944A (en) * | 2013-06-26 | 2013-09-25 | 电子科技大学 | Fake fingerprint detecting method based on SVM and sparse representation |
| CN103942540A (en) * | 2014-04-10 | 2014-07-23 | 杭州景联文科技有限公司 | False fingerprint detection algorithm based on curvelet texture analysis and SVM-KNN classification |
| CN104581143A (en) * | 2015-01-14 | 2015-04-29 | 宁波大学 | Reference-free three-dimensional picture quality objective evaluation method based on machine learning |
| CN105335707A (en) * | 2015-10-19 | 2016-02-17 | 广东欧珀移动通信有限公司 | Method and apparatus for acquiring fingerprint image to be identified, and mobile terminal |
| CN106021410A (en) * | 2016-05-12 | 2016-10-12 | 中国科学院软件研究所 | Source code annotation quality evaluation method based on machine learning |
| CN106709916A (en) * | 2017-01-19 | 2017-05-24 | 泰康保险集团股份有限公司 | Image quality assessment method and device |
| CN106780434A (en) * | 2016-11-15 | 2017-05-31 | 天津大学 | Underwater picture visual quality evaluation method |
-
2017
- 2017-11-10 CN CN201711103387.3A patent/CN107992800A/en active Pending
Patent Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP1603077A1 (en) * | 2003-03-07 | 2005-12-07 | Nippon Telegraph and Telephone Corporation | Biological image correlation device and correlation method thereof |
| US20050063562A1 (en) * | 2003-08-07 | 2005-03-24 | Brunk Hugh L. | Conveying fingerprint minutiae with digital watermarks |
| US20060120575A1 (en) * | 2004-12-04 | 2006-06-08 | Ahn Do S | Method and apparatus for classifying fingerprint image quality, and fingerprint image recognition system using the same |
| CN101303728A (en) * | 2008-07-01 | 2008-11-12 | 山东大学 | Method for identifying fingerprint facing image quality |
| CN103324944A (en) * | 2013-06-26 | 2013-09-25 | 电子科技大学 | Fake fingerprint detecting method based on SVM and sparse representation |
| CN103942540A (en) * | 2014-04-10 | 2014-07-23 | 杭州景联文科技有限公司 | False fingerprint detection algorithm based on curvelet texture analysis and SVM-KNN classification |
| CN104581143A (en) * | 2015-01-14 | 2015-04-29 | 宁波大学 | Reference-free three-dimensional picture quality objective evaluation method based on machine learning |
| CN105335707A (en) * | 2015-10-19 | 2016-02-17 | 广东欧珀移动通信有限公司 | Method and apparatus for acquiring fingerprint image to be identified, and mobile terminal |
| CN106021410A (en) * | 2016-05-12 | 2016-10-12 | 中国科学院软件研究所 | Source code annotation quality evaluation method based on machine learning |
| CN106780434A (en) * | 2016-11-15 | 2017-05-31 | 天津大学 | Underwater picture visual quality evaluation method |
| CN106709916A (en) * | 2017-01-19 | 2017-05-24 | 泰康保险集团股份有限公司 | Image quality assessment method and device |
Non-Patent Citations (4)
| Title |
|---|
| FERNANDO ALONSO-FERNANDEZ ETAL: "A Comparative Study of Fingerprint Image-Quality Estimation Methods", 《IEEE TRANSACTIONS ON INFORMATION FORENSICS AND SECURITY》 * |
| PEIJUN DU ETAL: "Hyperspectral remote sensing image classification based on the integration of support vector machine and random forest", 《2012 IEEE INTERNATIONAL GEOSCIENCE AND 《REMOTE SENSING SYMPOSIUM》 * |
| 张宇: ""指纹图像质量评价方法研究"", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
| 林海祥: "基于统计学习的无参考人脸图像质量评价", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (19)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109166091A (en) * | 2018-06-29 | 2019-01-08 | 中国移动通信集团江苏有限公司 | Detection method, device, equipment and medium |
| CN109166091B (en) * | 2018-06-29 | 2020-11-06 | 中国移动通信集团江苏有限公司 | Detection method, apparatus, device and medium |
| CN109376628A (en) * | 2018-10-10 | 2019-02-22 | 杭州登虹科技有限公司 | A kind of picture quality detection method, device and storage medium |
| CN110318327A (en) * | 2019-06-10 | 2019-10-11 | 长安大学 | A kind of surface evenness prediction technique based on random forest |
| CN110472518A (en) * | 2019-07-24 | 2019-11-19 | 杭州晟元数据安全技术股份有限公司 | A kind of fingerprint image quality judgment method based on full convolutional network |
| CN110472518B (en) * | 2019-07-24 | 2022-05-17 | 杭州晟元数据安全技术股份有限公司 | Fingerprint image quality judgment method based on full convolution network |
| CN111027629A (en) * | 2019-12-13 | 2020-04-17 | 国网山东省电力公司莱芜供电公司 | Power distribution network fault outage rate prediction method and system based on improved random forest |
| CN111027629B (en) * | 2019-12-13 | 2024-02-27 | 国网山东省电力公司莱芜供电公司 | Power distribution network fault power failure rate prediction method and system based on improved random forest |
| CN111127448B (en) * | 2019-12-26 | 2022-05-10 | 西南交通大学 | Method for detecting air spring fault based on isolated forest |
| CN111127448A (en) * | 2019-12-26 | 2020-05-08 | 西南交通大学 | Model and method for detecting air spring fault based on isolated forest |
| CN112115917A (en) * | 2020-09-29 | 2020-12-22 | 深圳市汇顶科技股份有限公司 | Fingerprint identification method, fingerprint identification device, electronic equipment and storage medium |
| CN112115917B (en) * | 2020-09-29 | 2024-05-28 | 深圳市汇顶科技股份有限公司 | Fingerprint identification method, fingerprint identification device, electronic equipment and storage medium |
| CN112287948A (en) * | 2020-10-30 | 2021-01-29 | 西安工程大学 | Sketch mark-based rapid edge detection method for extracting middle-level features |
| CN112287948B (en) * | 2020-10-30 | 2024-05-03 | 西安工程大学 | Sketch mark-based middle-level feature extraction rapid edge detection method |
| CN112926689A (en) * | 2021-03-31 | 2021-06-08 | 珠海格力电器股份有限公司 | Target positioning method and device, electronic equipment and storage medium |
| CN113361497A (en) * | 2021-08-09 | 2021-09-07 | 北京惠朗时代科技有限公司 | Intelligent tail box application method and device based on training sample fingerprint identification |
| CN113361497B (en) * | 2021-08-09 | 2021-12-07 | 北京惠朗时代科技有限公司 | Intelligent tail box application method and device based on training sample fingerprint identification |
| CN115342900A (en) * | 2022-08-15 | 2022-11-15 | 东北石油大学 | Laser self-mixing interference micro-vibration measurement method and system based on random forest |
| CN115342900B (en) * | 2022-08-15 | 2024-04-30 | 东北石油大学 | Random forest-based laser self-mixing interference micro-vibration measurement method and system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107992800A (en) | A kind of fingerprint image quality determination methods based on SVM and random forest | |
| CN106803247B (en) | Microangioma image identification method based on multistage screening convolutional neural network | |
| Lu et al. | Dense and sparse reconstruction error based saliency descriptor | |
| Kim et al. | Texture-based approach for text detection in images using support vector machines and continuously adaptive mean shift algorithm | |
| Mansanet et al. | Local deep neural networks for gender recognition | |
| Li et al. | Saliency detection via dense and sparse reconstruction | |
| Deng et al. | Saliency detection via a multiple self-weighted graph-based manifold ranking | |
| Tsai et al. | Image co-saliency detection and co-segmentation via progressive joint optimization | |
| Ferreira et al. | Behavior knowledge space-based fusion for copy–move forgery detection | |
| Esmaeili et al. | Fast-at: Fast automatic thumbnail generation using deep neural networks | |
| US20070242856A1 (en) | Object Recognition Method and Apparatus Therefor | |
| Huo et al. | Fast object-level change detection for VHR images | |
| CN110569782A (en) | Target detection method based on deep learning | |
| CN103996052A (en) | Three-dimensional face gender classification device and method based on three-dimensional point cloud | |
| US8027978B2 (en) | Image search method, apparatus, and program | |
| Zhang et al. | Saliency detection via extreme learning machine | |
| Huo et al. | Learning relationship for very high resolution image change detection | |
| Bai et al. | Principal pixel analysis and SVM for automatic image segmentation | |
| Inkeaw et al. | Recognition-based character segmentation for multi-level writing style | |
| CN111815582A (en) | Two-dimensional code area detection method for improving background prior and foreground prior | |
| Juang et al. | Stereo-camera-based object detection using fuzzy color histograms and a fuzzy classifier with depth and shape estimations | |
| Xia et al. | Fast template matching based on deformable best-buddies similarity measure | |
| Houtinezhad et al. | Writer-independent signature verification based on feature extraction fusion | |
| Elsayed et al. | Hand gesture recognition based on dimensionality reduction of histogram of oriented gradients | |
| Rohith et al. | Design of Deep Convolution Neural Networks for categorical signature classification of raw panchromatic satellite images |
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 |