Font Size: a A A

Study On Contactless Multiple Palm Feature Extraction Method

Posted on:2014-07-29Degree:DoctorType:Dissertation
Country:ChinaCandidate:W LiFull Text:PDF
GTID:1268330431952322Subject:Measuring and Testing Technology and Instruments
Abstract/Summary:PDF Full Text Request
Person’s palm contains rich palm vein and palmprint.Study has shown that thedistribution of palmprint is unique and stable.It can be used as the basis for authentication.Palm vein is the internal body characteristics, which has certain anti-falsification.Multi-modal palm recognition based on palmprint and palm vein can improve thereliability of the personal authentication. At the same time, the palm image can be acquiredby contactless acquisition mode, which is more hygienic and be adopted easily by peoplebecause of its non-intrusive.The above features make palm identification technology has agreat advantage in the access control system.Contrast to the contact acquisition mode, the direction, position and shooting distancebetween the palm and the imaging device is uncertain when collecting palm image withcontactless acquisition mode. There are two solutions. One is to normalize the palm image.Another is to extract the translation, rotation and scaling invariant hand feature. Thenormalization changes some properties of the original image that may affect the featureextraction. At the same time, the normalization process also increases the running time. Sothis thesis chooses the latter program.Main works and results of this thesis are as follows.(1) In order to select the stable reference point of palm, a palm localization methodbased on inscribed circle is proposed in this thesis. It chooses the circle center as the palmreference point. In order to ensure the inscribed circle is unique, the inscribed circle isdesigned to meet the following criterion: the inscribed circle is tangent to both side of thepalm contour and through the junction point of middle finger and ring finger. By thesymmetry of the circle, the perpendicular bisector that through any two points on theboundary will pass through the circle center and the tangent will be perpendicular to theradius that pass through the tangent point. According to these natures, this methoddetermines the centres and radiuses of inscribes circle by detecting a boundary point in the palm contour lines of the lateral index finger and lateral pinkie outside respectively. Finally,it judges the positioning stability by the standard deviation of gray difference surfacewithin inscribes circle. The more similar the image within inscribes circle, the higher thepositioning stability, and the smaller the standard deviation. The positioning method isverified based on self-build image database.The positioning stability of the proposedmethod is0.751and the average localization time is1116ms. The positioning stability oftraditional inscribes circle positioning method is0.837and the average localization timeis1555ms. Experimental results demonstrate that the proposed method can sresolve theproblem of palm positioning.Compared with traditional methods, the proposed method hashigher positioning accuracy but it need fewer localization time.(2) In this thesis, the relative positions of the vein crossover points and the inscribecircle center are constructed into a set rotation, translation, scaling invariant features.Firstly, the inscribed circle of palm is obtained. The relative radius and relative angle isdefined by the use of the intersection point inside circle and the circle center. Then, thetwo-dimensional feature vector space is established by the radius and angle parameters.And image is converted to a series of feature points in the feature space. Finally, similarityof the distribution structure of feature points in the feature vector space is taken as thematching basis. Use two methods to evaluate the effectiveness of the features. On the onehand, it is to analyze the stability of the features by match the original image and the imageafter rotation and scaling. Experimental results show that the rotation and scaling have noimpact on the constructed palm features. On the other hand, it is to analyze the uniquenessof the features. The performance of the proposed method is verified by self-build imagedatabase.When thumb naturally stretch and four fingers close, the equal error rate(EER) is0.96%.When five fingers naturally stretch, the equal error rate(EER) is4.91%.Experimental results show that the method can obtain good recognition effect. It hasinvariant to scaling, rotation and translation and fault tolerance of feature pointextraction.This method is suitable for palm identification instruments that need smallerdata storage space, because it can obtain better recognition results by a small number offeatures.(3) This thesis proposes a novel feature extraction method which can reflects thegeometric features of palmprint and palm vein without being affected by the scaling,rotation and translation. Firstly, several radiation segments are made between the inscribedcircle center and circumference. The reference direction is the direction of the circle center with the junction point of middle finger and ring finger. Meanwhile, the gradient value ofpixels in the feature line segment is calculated by creating template. The feature vectorspace is established by the relative radius of feature line segment’s centroids. Secondly, thepalm image database is established based on the practical application environment. Thefeature stability in different size of sub-template and recognition performance in differentnumber of feature line segments is analyzed. When the size of sub-template is7×7, thefeature stability is the best.When the number of feature line is sixty, the recognitionperformance is relatively stable and not any more significantly increased.Finally, use twomethods to evaluate the effectiveness of the features.1) The stability of the features.Onehundred and seventy hand images is rotated and scaled.The matching rate which theoriginal image and transformed images is greater than97%;2) The uniqueness of thefeatures.When the size of sub-template is equal to seven and the number of feature linesegments is sixty, the equal error rate(EER) is less than0.4%and feature extraction time is0.0019s. The experimental results show that the method can extract out stablecharacteristics whenever the hand image is scaled, rotated or translated. And computingspeed is faster.This method is suitable for palm identification device which has higherrequirements for real-time.
Keywords/Search Tags:Biometrics, Palm vein, Palmprint, Palm positioning, Feature Extracting
PDF Full Text Request
Related items