CN103324921B - A kind of mobile identification method based on interior finger band and mobile identification equipment thereof - Google Patents

Abstract

The invention provides a kind of unrestricted mobile identification method based on interior finger band, the method utilizes the photographic head of mobile device to gather hand images as collecting device, by network, collecting sample is transmitted to server end, and automatically carry out hand region detection, interior finger band zone location, interior finger striations characteristic extract and carry out feature with the interior finger band sample in data base and compare, thus realizing identification based on interior finger band biological characteristic.Present invention also offers a kind of mobile identification equipment realizing mobile identification method based on interior finger band, including: acquisition module, transport module, pretreatment module, processing module and decision-making module.Have and apply initiatively in mobile environment, it is suitable for the change of different background, illumination, attitude and viewpoint, for dislocation, there is certain tolerance, there is higher the match is successful rate is a kind of convenient, efficient and reliable identification system, there is in security fields the advantages such as preferable application prospect.

Description

A kind of mobile identification method based on interior finger band and mobile identification equipment thereof

Technical field

The present invention relates to a kind of mobile identification technology, particularly to a kind of mobile identification method based on interior finger band and Mobile identification equipment.

Background technology

Along with popularizing of the mobile device such as flat board, smart mobile phone, we are datatron Migong business on devices, Ge Renyin The work of the strict authentication of needs such as private business business is more and more frequent.But, traditional by arranging username and password The method identifying identity, does not have uniqueness, and easily passes out of mind, cracks or steal.With PIN(people's recognition code) no With, biological characteristic will not pass into silence, and loses or steals, it is impossible to is replicated easily or shares, therefore with biological characteristic as base The encryption recognition system that the identification system of plinth is relatively conventional has the biggest advantage.

There is the living creature characteristic recognition system that many is widely used, such as fingerprint, iris etc. now.Existing biological special Levy identification technology to have and following carry a little:

First, most biological identification technology depends on specialty collecting device and builds a controlled sample collection ring Accuracy of identification, to simplify recognizer, is improved in border.Even if part uses common camera, still environment is existed certain requirement, know Other precision is the highest.Therefore, in the environment of movement is uncontrollable, prior art is difficult to the recognition effect obtained.

Second, the recognition methods of Behavior-based control feature, such as gait, sound, signature and keystroke typewriting etc., accuracy of identification is not High and the most imitated.Therefore, for the hardware condition of existing mobile device, general consider to utilize photographic head to gather palmmprint, interior Refer to that the image informations such as band are as basis of characterization.

In Human and machine recognition of faces:a survey(people and the face recognition of machine: Investigation) in, it is proposed that a movable recognition system based on face, but, this system, in order to realize reliability, needs a bag Huge data base containing the shooting sample under the conditions of a large amount of different illumination and posture.And for Palm Print Recognition System, they are often The most high-resolution picture is needed to carry out identification to obtain enough structural informations, but when posture changes, high score Resolution can cause again picture substantially to be affected by projective transformation simultaneously.

At Illumination ratio image:synthesizing and recognition with varying Illuminations(illumination quotient images: synthesize under variable illumination and identify) in mention, compared with sound, face and palmmprint, Refer to that band is more suitable for mobile biological recognition system due to its physical feature.Finger band is except there being smooth surface, the most in distress With the abundant structural information imitated, again owing to it is in a region the least, even may be used so its projective transformation is the least To ignore.It is to say, utilize under the conditions of low resolution, to refer to that band is identified obtaining more higher than using personal recognition Accuracy rate.

In the use having pointed out at present refers to the method that band carries out bio-identification, solve posture the most very well and change Becoming the identification difficulty brought with illumination condition, they need use special installation and carry out the figure shot in ideal circumstances mostly Sheet, and do not account for the impact caused when condition is inconsistent when.

Document A Biometric Identification System Based on Eigenpalm and The Eigenfinger Features(feature based palm and feature refer to the living creature characteristic recognition system of feature) and A multi- Matcher system based on knuckle-based features(many adapters based on articulations digitorum manus feature system) The method of middle proposition needs to use scanning device to obtain picture, document Online finger-knuckle-print Verification for personal authentication(is for the finger band authenticating party online of person identification Method) in method need also exist for special system.

Document Palmprint Recognition across Different Devices(palmmprint on different devices Recognition methods) although in method be to use mobile device to gather palm picture, but it requires that picture background must be completely black, And illumination to accomplish the most uniform.

Summary of the invention

The primary and foremost purpose of the present invention is that the shortcoming overcoming prior art is with not enough, it is provided that a kind of shifting based on interior finger band Dynamic recognition methods, the method breaches to the restriction of environmental condition when gathering image in existing method, thus is more applicable for moving Dynamic equipment.

Another object of the present invention is to the shortcoming overcoming prior art with not enough, it is provided that a kind of realization is based on interior finger band The mobile identification equipment of mobile identification method, this equipment mainly includes the detection of image acquisition, hand, area-of-interest (ROI) Location, interior finger striations characteristic extract and mate etc. function.

The primary and foremost purpose of the present invention is achieved through the following technical solutions: a kind of mobile identification method based on interior finger band, Comprise the following steps:

T1: gather image, user in a mobile environment, uses the hand figure of the photographic head shooting of configuration in mobile device Picture.

T2: hand detects, through skin model based on mixed Gauss model and morphologic expansion, the method for corrosion After process, from source images, extract complete and accurate hand profile.

T3:ROI positions, and utilizes finger datum mark localization method and Radon projection to enter one from the hand images that T2 extracts Step is partitioned into four area-of-interests.

T4: interior finger striations characteristic extracts, and extracts the characteristic information of ROI region based on Competition coding method, including Orientation map(directional diagram) and energy map(energy diagram).

T5: interior finger striations characteristic mates, two spies that will obtain in above-mentioned steps T4 based on region histogram statistical method Levy map to carry out mating and obtain final matching result with the characteristic in server.

The detection of T2: hand mainly includes coarse localization (T21) and is accurately positioned (T22) two steps:

T21: coarse localization, with pre-prepd typical case's broca scale picture as training set, utilizes the expectation after improving maximum Change (EM) algorithm, carry out iterative computation likelihood function by existing data, be allowed to converge on certain optimal value, thus automatically obtain The parameter of Gaussian mixture model.Judging whether each pixel of the image gathered meets certain gaussian kernel, if meeting, judging For hand skin region, otherwise it is judged as background area, thus obtains rough hand images.

T22: be accurately positioned, utilizes morphological method to process rough hand images obtained above, can remove the inside Duck eye and noise, thus obtain complete and accurate hand images and contour line.

T3: after extracting complete accurate hand images, be partitioned into four area-of-interests from image.Main point For two steps (T31 and T32):

The location of T31: finger and segmentation.We utilize the distance of point on contour line to close in hand images obtained above It is positioning datum point: five finger tip points and four finger valley points.Estimate often refer to the axis of finger and extract based on these datum marks The region of finger.Less, so we only position remaining four finger owing to referring in thumb that band comprises information.

T32: the finger areas image split by T31 is rotated to level by axis, and all samples are cut to unified Size.Sample is done Radon projection, obtains obvious two peak regions, be interior finger band region.Recycling low pass Filtering obtains particular location coordinate, obtains final ROI region.

T4: utilize method based on Competition coding, extracts for identification from the region of interest area image that T3 obtains Feature.Method is specifically described as: we utilize Gabor filtering to catch the directional information of picture of publishing picture from image.By obtaining not After equidirectional Gabor filtering, the source images of the Gabor verification area-of-interest in each direction is utilized to carry out convolution operation also Finally giving n and open response value image, the most corresponding n different directions, the direction comprising minimum response value will be elected to be (being dominant) Direction, and form a width and represent the directional diagram (orientation map) of each pixel principal direction；Meanwhile, this is Little response value can be stored into another width figure, generates describe each pixel principal direction through the negating of numerical value, quantization operation The energy diagram (energy map) of weights.Final orientation map and energy map represents that the structure of image is special jointly Reference ceases.

T5: feature matching method based on region histogram statistics, mainly includes two steps (T51 and T52):

T51: we are obtaining orientation map(directional diagram from T4) and energy map(energy diagram) carry out local Statistics with histogram.We are divided into many overlapping regions source images, are called block.Each piece by some nonoverlapping unit Lattice form.Each cell correspond to a rectangular histogram formed by interior pixels by Nearest Neighbor with Weighted Voting.The throwing of each grid Ticket result, i.e. corresponding rectangular histogram constitutes the feature of each block, and the feature of each block constitutes again the feature of whole image Vector, for final identification.

T52: calculate the Euclidean distance of the characteristic vector stored in above-mentioned characteristic vector to be matched and background data base.This The bright method by machine learning obtains a threshold value, if Euclidean distance is less than threshold value, then characteristic matching success, completes user The certification of identity.Owing to have employed the method for statistics with histogram in block, the matching algorithm of the present invention is within the specific limits to two Scheme the position relationship between block to be matched the most sensitive, so existing a range of flat between sample to be matched When moving dislocation, still can effectively show the similarity degree of two figures, so the present invention has certain tolerance to dislocation.

Another object of the present invention is achieved through the following technical solutions: a kind of realize mobile identification side based on interior finger band The mobile identification equipment of method, mainly includes with lower module:

Acquisition module, for shooting the hand images of user；

Transport module, for user's hand images of being photographed from described collecting unit by network transmission to server End；

Pretreatment module, for detecting hand region from user's hand images of described server end, and positions interior finger Band region；

Processing module, refers to the result in band region, for extracting and mating user's based on described pretreatment unit location Interior finger band information；

Decision-making module, the result of interior finger band information based on described processing unit coupling user, be used for determining checking or The result identified.

The equipment that described acquisition module uses is the photographic head being equipped with in mobile device, the direction of photographic head and hands the five fingers exhibition Opening formed plane vertical, the shake of photographic head is in the range of 20 degree.

When acquisition module gathers, hand and background thereof have the feature that

Background there are differences with the color of hand skin；

Bright and clear, there is no strong shadow；

The five fingers are the most open and the most flat, separately；

Described pretreatment module includes hand region detector unit and the positioning unit of interior finger band, and described hand region is examined Surveying unit uses skin model based on mixed Gauss model to extract hand region, and described interior finger band positioning unit utilizes hands The Picking up geometry information finger areas of contouring, does Radon projection to finger areas, refers to band region in extracting.

Described processing module includes feature extraction unit and the matching unit in interior finger band region,

The character representation method of the Competition coding described in extraction unit employing of described feature；Unit input is described sense Interest regions, is output as the directional diagram described in claim 3 and energy diagram；

The method based on partial statistics described in matching unit employing of described feature；Unit input is described directional diagram And energy diagram, output is the Euclidean distance of the characteristic vector of image in the data base that this input picture and carrying out therewith mates.

Described decision-making module is that matching result based on a whole hands does decision-making, described based on multizone comprehensive certainly Plan method, comprises both of which: Validation Mode and recognition mode；

Described Validation Mode is by man-to-man coupling, it is judged that in target image and data base, whether image is same Hands；

Described recognition mode is the coupling of one-to-many, finds the hands mated most from the data base of hands, if decision-making energy Value less than threshold value, then refers to not exist in band data base the picture of the hands of detected user in judging.It is employed herein weighting The method of meansigma methods calculates decision-making energy value, and due to little finger band, to generally comprise information less, so weights are relatively low, eating Finger, middle finger, the third finger, the weights of little finger of toe are respectively set as 0.3, and 0.3,0.3,0.1, therefore, total decision-making energy value T=0.3* (T2 + T3+T4)+0.1*T5, wherein T2, T3, T4, T5 represent the energy value of forefinger, middle finger, the third finger, little finger of toe respectively.

The operation principle of the present invention: the present invention utilizes the camera collection hand images of mobile device, utilizes based on mixing The skin model of Gauss model extracts hand region, by analyzing the interior district referring to band, geometric properties location of hand contour line Territory, utilizes Gabor filtering technique based on Competition coding refer to striations characteristic in extracting and use region histogram to mate, It is achieved thereby that identification based on interior finger band biological characteristic.

The present invention has such advantages as relative to prior art and effect:

1, this method is by utilizing feature interior relatively stable finger band and image processing method, effectively reduces light According to user posture on identify image impact, overcome at document Human and machine recognition of Faces:a survey(people and the face recognition of machine: investigation) in need one to comprise a large amount of different illumination and posture under the conditions of The shortcoming of huge data base of shooting sample.

2, at document Illumination ratio image:synthesizing and recognition with Varying illuminations(illumination quotient images: synthesize under variable illumination and identify) in mention, with sound, face and Palmmprint is compared, and refers to that band is more suitable for mobile biological recognition system due to its physical feature.Refer to that band is except there being smooth table Face, is also difficult to the abundant structural information imitated, again owing to it is in a region the least, so its projective transformation is very Little even can ignore.It is to say, this method utilization refers to that band is identified obtaining than use under the conditions of low resolution The higher accuracy rate of personal recognition.

3, this method has only to the mobile device with common camera and can realize, and overcomes document ABiometric Identification System Based on Eigenpalm and Eigenfinger Features(feature based is slapped With the living creature characteristic recognition system that feature refers to feature), A multi-matcher system based on knuckle-based Features(many adapters based on articulations digitorum manus feature system) and document Online finger-knuckle-print Verification for personal authentication(is for the finger band authenticating party online of person identification Method) in need the extra restriction of special installation.

4 and document Palmprint Recognition across Different Devices(on different devices Palm grain identification method) in require that picture background must be completely black, and illumination to accomplish that uniform method is compared as best one can, this method Refer to band and image processing method in utilizing feature significantly, effectively reduce the requirement of picture background and photoenvironment.

5, compared with prior art, present method be advantageous in that the condition limit that can reduce largely when gathering image System.Traditional method must be identified the collection of image by specific assisted acquisition instrument, or when gathering image pair Surrounding requires stricter.And our method refers to striations characteristic in make use of more significantly, it is only necessary to set with mobile The photographic head of standby upper configuration is as sampling instrument.Extract during feature further through various methods reduce as far as possible image by Exposure under different situations, rotate, the mobile change caused, thus different environmental conditions is had stronger robustness.

Accompanying drawing explanation

Fig. 1 is skin model based on mixed Gauss model.

Fig. 2 is the workflow diagram of the on-site identification process of the present invention.

Fig. 3 a is to obtain coarse hands contour images respectively through after skin model detection.

Fig. 3 b is the more complete accurate hands contour images obtained after expansion, etch state process.

Fig. 4 a is that finger datum mark positions schematic diagram.

Fig. 4 b is the schematic diagram of location, finger axis.

Fig. 4 c is the schematic diagram after finger segmentation.

Fig. 5 is the schematic diagram that the area-of-interest (ROI region) of interior finger band based on RADON projection positions.

Fig. 6 a is with the method schematic diagram of localized mass sector scanning directional diagram.

Fig. 6 b is block structure schematic diagram.

Fig. 6 c is bicubic interpolation schematic diagram.

Fig. 7 is the equipment structure chart of the present invention.

Fig. 8 is the flow chart of hand images pretreatment module.

Fig. 9 is ROC based on ecotopia and mobile environment figure.

Detailed description of the invention

Below in conjunction with embodiment and accompanying drawing, the present invention is described in further detail, but embodiments of the present invention do not limit In this.

Embodiment

The whole identification process of the present invention includes two parts: preliminary preparation and field conduct.

Preliminary preparation includes two parts: set up hand images data base and skin model.

Set up hand images data base, gather the hand images of targeted customer.Collecting device is common camera, and it gathers Condition mainly comprises following feature: background there are differences with the color of hand skin；Light is more sufficient, does not has reinforcing YIN-essence Shadow；The five fingers are the most open and the most flat, separately；The plane of the positive opponent in direction of photographic head, shake is in the range of 20 degree.

Setting up skin model, the extraction typical case's area of skin color from hand images data base, as training set, utilizes and improves After expectation-maximization algorithm, automatically obtain the parameter of Gaussian mixture model.As shown in Figure 1, it is shown that represent the 8 of skin model The distribution of individual gaussian kernel.

The flow process of field conduct is as shown in Figure 2:

#1 gathers user's hand images；

Image is uploaded onto the server by #2；

#3 hand extracts；

#4 finger locating and segmentation；

The area-of-interest of band is referred in #5 location；

Refer in #6 that striations characteristic extracts；

The feature extracted is mated in data base by #7；

#8 integrated decision-making；

#9 server returns matching result to mobile terminal；

In flow chart, the ins and outs of concrete each step are as follows:

#1 gathers user's hand images: user uses the photographic head shooting hand images of configuration in mobile device, and it gathers The feature of condition is as follows: background there are differences with the color of hand skin；Light is more sufficient, does not has strong shadow；The five fingers are natural Open and flat, separately；The plane of the positive opponent in direction of photographic head, shake is in the range of 20 degree；Making palm include, that finger occupies substantially is whole Individual image frame.

Image is uploaded onto the server by #2: mobile device accesses the Internet, by collect by 3G or this WIFI network Hand images uploads to background server.

#3 hand extracts: carries out the coarse localization of hand region based on the skin model being obtained ahead of time, is specifically described as: sentences On the disconnected image gathered, whether each pixel meets certain gaussian kernel, if meeting, being judged as hand skin region, otherwise sentencing Break as background area, thus obtain rough hand images, as shown in Figure 3 a.On this basis, morphological method is utilized to remove Duck eye in above-mentioned rough result and noise, thus obtain complete and accurate hand images and contour line, as shown in Figure 3 b.

#4 finger locating and segmentation: based on hand region obtained above, we utilize the distance relation of point on contour line Positioning datum point: five finger tip points and four finger valley points.As shown in fig. 4 a, we are from the beginning of an end points P, along contour line by Point statistics and the distance of some P, (such as increasing from P to T5 front distance when distance variation tendency occurs and substantially changes always, arriving Start after T5 to reduce), choose current putting as characteristic point, so we can obtain five finger tip points of T1-5, B1, B3, B4, B5 It it is a finger valley point.Moreover, we find the some B2 nearest with B3 on the contour line between T2 and B1, in like manner find a B6 As new characteristic point, so we determine the Position Approximate of finger the most substantially.

As a example by middle finger, as shown in Figure 4 b, we, by two curve trisections of B3T3, B4T3, obtain M1, M2 ', M3, M4 ' Four trisection points, then from M2 point start travel through each point to upper and lower both direction along contour line, calculate this point with The distance of M1, obtains nearest one M2, in like manner obtains M4, connects M1M2, M3M4, takes the line D2V2 of these two sections of emphasis, i.e. Axis for this finger.Do rectangle frame with this axis for axle, i.e. can get finger areas.

Less owing to referring in thumb that band comprises information, we only position remaining four finger, as illustrated in fig. 4 c.

The area-of-interest of band is referred to: finger areas previous step split rotates to level, to often in #5 location One finger areas does Radon projection, and each finger areas can obtain two peak values, each finger areas is utilized low pass filtered Ripple obtains the interior finger band region that determines, and Fig. 5 refers to the extraction of band in representing in a finger areas.

Refer in #6 that striations characteristic extracts: conventional extracting method can lose efficacy in the case of color change is smoother, In order to solve this problem, the present invention use Competition coding method from ROI region image internally refer to band structural information carry out Extract.In the present invention, Gabor filter is used to extract the directional information referring to band.The present invention passes through Gabor kernel at n In individual angle, image pixel being carried out process of convolution, convolution kernel is calculated by a Gabor function based on neuro physiology Go out.This function is:

X '=cos θ x+sin θ y,

$\mathrm{\κ}=\sqrt{2\ln }\left(\frac{2\mathrm{\φ}+1}{2\mathrm{\φ}-1}\right),$

$K=\frac{\sqrt{2\mathrm{\π}}}{\mathrm{\κ}\·\mathrm{\λ}};$

Wherein θ is the direction of small echo, typically takes 18, and the most every 10 degree take one and represent direction；σ is the standard of Gaussian profile Difference；λ andIt is frequency and the phase place of SIN function respectively.After convolution, obtain 18 response value images, then contrast each Individual computing unit sensitivity on 18 directions, response value is the least, represents that directivity is the strongest, the direction that response value is minimum Be set to the principal direction of this computing unit, the principal direction of all computing units stored, formed describe on image each as The directional diagram (orientation map) of vegetarian refreshments principal direction.Meanwhile, the response value of this minimum can be stored into another width Figure, by the data inside the diagram are deducted minima, inverted, and quantify to operations such as [0,1.0], generate and represent The energy diagram (energy map) of the weights of each pixel principal direction.The weights of energy map are the biggest, represent this pixel The lines embodied in this principal direction are the strongest.Final orientation map and energy map describes jointly The structure feature information of image.

The feature extracted is mated in data base by #7: obtain the structural information of interior finger band After (orientation map, energy map), the method that the present invention uses region histogram to add up achieves tolerance dislocation Image spatial feature mates.Specifically can be described as:

First, define a block region, each piece of region comprises 2x2 subelement, and each subelement comprises again 4X4 picture Vegetarian refreshments, as shown in Figure 6 b.

Second, from left to right, from top to bottom, with the orientation map(side described in a block sector scanning right 6 To figure).The step-length from left to right moved is 4 pixels, and the step-length moved from top to bottom is 4 pixels, shown in institute Fig. 6 a.

3rd, in each block region, the region of corresponding orientation map (directional diagram) is carried out based on directly The statistics of side's figure.For each subelement, respective energy value on 6 directions of statistics respectively.Calculating each pixel The when of to the contribution of 6 different directions of different subelements, it is necessary to according to locus and the orientation angle of this pixel Weights are carried out bicubic interpolation, as fig. 6 c.Here weights refer to that this pixel is corresponding at energy map (energy diagram) On value.Thus, each subelement comprises the characteristic vector of one 6 dimension, also implies that, each block comprises the spy of one 24 dimension Levy vector.The characteristic vector in each block region is combined, is the formation of describing the characteristic vector of entire image.

4th, calculate two images similarity structurally, it is simply that calculate the distance of two width image feature vectors, use Euclidean distance computational methods.By substantial amounts of experimental data, we predict a reliable distance as threshold value, based on this threshold Value refers to whether band mates in judging two.Threshold value described here is an empirical value based on great many of experiments test.

#8 integrated decision-making: in order to obtain more structurally sound coupling structure, the present invention uses integrated decision-making based on multizone Method, specifically can be described as: for each hands, and the region of coupling includes Section three of four finger fingers in addition to thumb Interior finger band a, say, that hands comprises the subregion of four couplings.The matching result of four sub regions is simply by the presence of two Or above result is judged as coupling, then it is assumed that the image of two corresponding width handss is from the same hand.Decision-making energy value is The weighted average of the coupling energy value in the interior finger band region matched.

#9 server returns matching result to mobile terminal: the result of coupling is passed to the movement of correspondence by server by network End, matching result shows in mobile terminal.

Based on the equipment needed for above flow process, the unrestricted mobile identification method based on interior finger band proposed in the present invention Comprise 5 modules, as shown in Figure 7.Equipment is by acquisition module, transport module, pretreatment module, processing module and decision-making module group Become.

Acquisition module is responsible for shooting the hand images of user；

User's hand images that transport module is responsible for being photographed from described collecting unit by network transmission is to server end；

Pretreatment module is responsible for detecting finger band ROI region in hand region, and location from the image that collection comes；

The interior finger band ROI region that processing module is responsible for obtaining carries out feature extraction and matching；

Decision-making module is responsible for determining checking or the result identified based on matched data.

Pretreatment module includes hand region detector unit and the positioning unit of interior finger band, it is characterised in that: described hands Portion's region detection unit uses skin model based on mixed Gauss model to extract hand region, and described interior finger band location is single Unit utilizes the Picking up geometry information finger areas of hand profile, and finger areas is done Radon projection, refers to band ROI district in extracting Territory, idiographic flow is as shown in Figure 8.

Processing module includes feature extraction unit and the matching unit in interior finger band region, and wherein, feature extraction unit is adopted By the character representation method of Competition coding, input as interior finger band ROI region image, be output as directional diagram (orientation And energy diagram (energy map) map)；Characteristic matching unit uses method based on partial statistics, inputs as directional diagram (orientation map) and energy diagram (energy map), output is the characteristic vector of image in target image and data base Euclidean distance.

Decision-making module includes verifying (verification) unit and identifying (identification) unit.Decision-making module Being Comprehensive Model based on multizone, each hands mates 4 each ROI region, is middle forefinger, middle finger, the third finger, little respectively The third knuckle referred to.Authentication unit is man-to-man coupling, it is judged that in target image and data base, whether image is the same hand； Recognition unit is the coupling of one-to-many, finds the hands mated most from the data base of hands, the hands that i.e. decision-making energy value is the highest, if Decision-making energy value is less than threshold value, then it is assumed that there is not the picture of the same hand in data base.

For the method verifying the present invention, we and take under the consistent ecotopia of illumination respectively in a mobile environment Obtaining sample and carry out coupling experiment, is the experimental result of the present invention as shown in table 1 and Fig. 9:

Sample type	Sample size	Correct coupling	Erroneous matching	Rate that the match is successful
					Ecotopia	1000	992	8	99.2%
Mobile environment	500	485	15	97.0%

Table 1

Under mobile environment, 22-> 15,

From experimental result it can be seen that present invention identification in a mobile environment can also keep preferable matching effect, institute Method before comparing, our invention is more suitable for being applied in mobile device.

That Fig. 9 represents is present invention Receiver operating curve in the environment of ideal and under mobile environment, wherein GAR represents correct receptance, and FAR represents false acceptance rate, it can be seen that our method is permissible in ideal circumstances In the case of the lowest FAR, obtaining the highest GAR value, comprehensive matching rate is more than 99%.Even if in a mobile environment, we also may be used To be easily reached 97% the match is successful rate, this indicates that the performance of the present invention meets use demand.

Above-described embodiment is the present invention preferably embodiment, but embodiments of the present invention are not by above-described embodiment Limit, the change made under other any spirit without departing from the present invention and principle, modify, substitute, combine, simplify, All should be the substitute mode of equivalence, within being included in protection scope of the present invention.

Claims (7)

1. a mobile identification method based on interior finger band, it is characterised in that comprise the following steps:

Step 1, set up with indoor finger band data base；

Step 2, automatically detect hand region；

Band region is referred in step 3, location；

Step 4, illumination, attitude and viewpoint are had the character representation of robustness；

Step 5, dislocation had the characteristic matching of tolerance；

Step 6, based on multizone integrated decision-making；

In described step 4, the described character representation to illumination, attitude and viewpoint with robustness refers to spy based on Competition coding Levying method for expressing, described character representation method based on Competition coding comprises the following steps:

A, each pixel for interior finger band region, calculate the energy value that the Gabor in n direction filters, and energy value is That little direction is set to the principal direction of this pixel；Described n direction refers to 360 degree to be divided into n interval, and n is for being more than The integer of 18；

Refer in B, a width that band image, through Competition coding, exports the characteristic image of the two same resolution of width；Represent each pixel The image of some principal direction, referred to as directional diagram；Represent the weights of each pixel principal direction, referred to as energy diagram；

In described step 5, the described feature matching method to dislocation with tolerance refers to characteristic matching based on partial statistics Method, described feature matching method based on partial statistics comprises the following steps:

(1) one block region of definition, comprises a square subelement of c, and each subelement comprises again a square pixel of S；

(2) on described directional diagram, from left to right, from top to bottom, it is scanned with a block region；From left to right move Step-length is d1, and the step-length moved from top to bottom is d2, and step-length d1 and d2 must assure that between block region, there is weight in the region of at least 1/3 Folded；Scan each time, the space that block region is corresponding is carried out based on histogrammic statistics；Rectangular histogram for block region is united Meter；

In described step (2), the statistical method that the described rectangular histogram for block region carries out adding up is:

1. each subelement being added up energy value on m direction respectively, m is necessarily less than described n；

When 2. calculating each pixel to the contribution of m different directions of different subelements, according to the space bit of this pixel Putting and orientation angle carries out bicubic interpolation to weights, described weights are the corresponding value on energy diagram of this pixel；

3. the feature vector dimension of each subelement is m, then the feature vector dimension in each block region be c square and m Product；

4. the characteristic vector in each block region is combined, and forms the characteristic vector describing entire image；

5. in employing Euclidean distance computational methods calculate target image and data base, the distance of image feature vector, calculates to be matched Image between similarity in structure；

In described step 1, described is to use the photographic head shooting user's finger in mobile device to move with indoor finger band data base Hand images time dynamic and set up；Gather user's hand images: user uses the photographic head shooting hands of configuration in mobile device Portion's image, acquisition condition is as follows: background there are differences with the color of hand skin；Light is more sufficient, does not has strong shadow；The five fingers The most open and the most flat, separately；The plane of the positive opponent in direction of photographic head, shake is in the range of 20 degree；

In described step 3, the method referring to band region in described location, comprise the following steps:

The first step: set up the mixed Gauss model generated by broca scale picture as training set, improves expectation-maximization algorithm and obtains The parameter of each Gauss distribution in described mixed Gauss model, and set up skin model；

Second step: each pixel color value of hand images of described shooting is substituted into each height of the skin model having built up In this distribution, if pixel value meets Gauss distribution, then it is determined as hand region, is otherwise background area；

3rd step: be filtered the image of described shooting by the method for the expansion in morphology and corrosion, eliminates scrappy region, Obtain complete hand contour line；

4th step: with the distance relation positioning datum point of point on described complete hand contour line；

5th step: determine the position of every finger in the image of described shooting by the geometrical relationship of described datum mark Yu described contour line Put and axis；

6th step: obtain area-of-interest with the image of shooting described in Radon Projection Analysis.

Mobile identification method based on interior finger band the most according to claim 1, it is characterised in that in described step 6, institute The method stating integrated decision-making based on multizone comprises the following steps:

I, set the detected forefinger of hands, middle finger, the third finger, these four fingers of little finger of toe interior finger band as the sub-district of coupling Territory；

II, every sub regions is each carried out described character representation based on Competition coding and feature based on partial statistics Join；

III, the subregion of these four fingers of forefinger, middle finger, the third finger and little finger of toe is compared with the data with indoor finger band data base Relatively, if there is the fingers of two or more than two with during with the data match of indoor finger band data base, then quilt is judged The image of hands of detection matches with the image of the hands mated in data base；

IV, decision-making energy value be these four fingers of forefinger, middle finger, the third finger and little finger of toe with by the data of indoor finger band data base The weighted mean of the coupling energy value in the interior finger band region matched.

3. realize a mobile identification equipment for mobile identification method based on interior finger band described in claim 1, its feature It is, including:

Acquisition module, for shooting the hand images of user；

Transport module, for user's hand images of being photographed from described acquisition module by network transmission to server end；

Pretreatment module, for detecting hand region from user's hand images of described server end, and positions interior finger band Region；

Processing module, based on the result in finger band region in described pretreatment module location, for extracting and mating the interior of user Refer to band information；

Decision-making module, the result of interior finger band information based on described processing module coupling user, it is used for determining checking or identifying Result.

Mobile identification equipment the most according to claim 3, it is characterised in that: the equipment that described acquisition module uses is mobile The photographic head being equipped with on equipment, the plane that the direction of described photographic head launches to be formed with the five fingers of hands is vertical.

Mobile identification equipment the most according to claim 3, it is characterised in that described pretreatment module includes that hand region is examined Surveying unit and the positioning unit of interior finger band, described hand region detector unit uses skin model based on mixed Gauss model Extracting hand region, described interior finger band positioning unit utilizes the Picking up geometry information finger areas of hand profile, to finger Radon projection is done in region, refers to band region in extracting.

Mobile identification equipment the most according to claim 3, it is characterised in that: described processing module includes interior finger band region Feature extraction unit and characteristic matching unit,

The character representation method of the Competition coding described in the employing of described feature extraction unit；Feature extraction unit input is for described Area-of-interest, is output as described directional diagram and energy diagram；

Method based on partial statistics described in the employing of described characteristic matching unit；The input of characteristic matching unit is described direction Figure and energy diagram, output is the Euclidean distance of the characteristic vector of image in the data base that input picture and carrying out therewith mates.

Mobile identification equipment the most according to claim 3, it is characterised in that: described decision-making module is coupling based on hands knot Fruit is cooked decision-making, described Synthetic Decision Method based on multizone, comprises both of which: Validation Mode and recognition mode；

Described Validation Mode is by man-to-man coupling, it may be assumed that judge in image and the interior finger band data base of detected hands The image of hands whether be the same hand；

Described recognition mode is the coupling of one-to-many, it may be assumed that find the image of the hands mated most from the interior finger band data base of hands, If decision-making energy value is less than threshold value, then refer to band data base does not exist the picture of the hands of detected user in judging.