Research On Hand Shape Recognition Method In Non-ideal Conditions

In digitized and networked current society, the identity recognition has been gained moreand more attention in order to guarantee the security of the information. As a result, itsindustry scale is increasingly expanded. As one of the biometric features, hand shape featurehas its own characteristics. Firstly, hand shape features are difficult to be obtained due tohands are usually half clenched in natural situation. Therefore, it has certain confidentialityand is difficult to be stolen for copying. Secondly, when the hand is in a flat state, the difficultyof acquisition and the requirement of image quality for hand shape features are lower than thatof iris and fingerprint because their characteristics are based on the texture information.Consequently, hand shape features are easy to be obtained by contactless manner so as toreduce the infection of bacteria and improve the acceptability of the users. Thirdly, comparedwith other textural features, the hand shape feature has a faster processing speed because of itseasier extraction methods. Therefore, it is suitable for authentication systems or can be mergedtogether with other features although the feature doesn't have the high uniqueness.Currently, for the demand of the practical application, the research of hand shape ismainly focused on finding a hand shape recognition method with robustness in contactlessacquisition condition. When the hand is separated with any parts of the acquisition device, thephenomena named as non-ideal phenomenon in this paper such as translation, rotation, scalingand inclination deformation are happened due to the difference feeling of placing hands fordifferent users. These phenomena will reduce the accuracy of hand shape features recognitionin acquired images and bring the additional error of the size for hand shape features. The mainresearch content of this paper is the hand shape recognition of non-ideal images with simplebackground interference. The specific work can be divided into the following aspects.(1) Contrapose the question of positional accuracy of finger features influenced by handtranslation, rotation, deformation and the difference of the opening degree of the fingers in theimages caused by the non-ideal phenomena, the hand area figure is firstly analyzed byclustering of the YCbCgCr color space model, then the hand shape contour image is extracted.Finger length feature points are located by using curvature method without influence of palmposition. In order to avoid the width mutations, the width of finger is obtained by calculating the average values for different position of midline of the finger. Compared with othermethods by the recognition performance after normalized each hand feature, the correctrecognition rate (CRR) of the finger length is increased by3.4%than the linear fitting method,and the CRR of the finger width is increased by1.5%than the mid-point method. So it isproved that the locating methods of the finger features has better robustness in the non-idealconditions.(2) Contrapose the question of positional accuracy of wrist point and the palm widthpoints caused by the flexible rotation of the wrist and the unfixed opening degree of thefingers as using the contactless acquisition, the candidate areas of the up and down wrist areaare firstly defined when locating the wrist point, then the wrist point can be obtained from theconfirmation of both sides of wrist point located by several maximum curvature points fromeach area. When locating the palm width point, the palm width is extracted refer to thepalmprint line on the little finger edge by considering the stability of the thumb root point islow. Compared with the Triangle-area method and Distance method after normalized thevalues of the palm features, the relative standard deviation of the palm length and palm widthare decreased by1.63and0.61, respectively, which are proved the high accuracy of theposition method under the non-ideal condition. In addition, the accuracy of hand shaperecognition also can be increased by increasing the palm features in the vector.(3) Contrapose the question of extra error of the hand shape feature value resulted by thescaling and inclination deformation of palm caused by the unfixed distance and angle betweenthe camera and hand due to the unfixed position of the hand location in the contactlessacquisition, two kinds of deformation phenomena are analyzed by using image optical modelin this paper. It is demonstrated by theoretical deduction that deformation caused byinclination deformation is much less than scaling. Therefore, after choosing the relativefeatures to avoid the interference of scaling, the stability and the recognition performance ofthe relative features after deformation are analyzed. The results are proved that the CRR of thedefined relative features in this paper can reach to88.93%, i.e., the relative features can be asstable features used to identify in small scope sample, and it can achieve effectiveidentification.(4) Contrapose the question of reduction of recognition performance caused by thresholdclassification because that high correlation among the different features make the proportion of the important features with inhomogeneity remarkable difference in the featuresummation become smaller in recognition process of hand shape feature, a information gainand correlation coefficient are presented for analyzing and evaluating the distinction and thecorrelation of the classification, and then the method of hand shape feature selection isoptimized by comprehensively analyzing the two parameters. The method can get rid of boththe low distinction features and the high relevance features, reducing the features by23as thereduction of recognition performance is0.7%. Moreover, the complexity of the method islower because it dose not get the subset one by one. The dimension of the optimized subset isreduced, so the recognition speed will be faster. The method can achieve the effect of effectiveidentification with less number of features. It is useful to present the advantage of hand shape.It can be easily merged with other features.