Behavior based authentication for touch screen devices

Abstract

A method, system, and one or more computer-readable storage media for behavior based authentication for touch screen devices are provided herein. The method includes acquiring a number of training samples corresponding to a first action performed on a touch screen of a touch screen device, wherein the first action includes an input of a signature or a gesture by a legitimate user. The method also includes generating a user behavior model based on the training samples and acquiring a test sample corresponding to a second action performed on the touch screen, wherein the second action includes an input of the signature or the gesture by a user. The method further includes classifying the test sample based on the user behavior model, wherein classifying the test sample includes determining whether the user is the legitimate user or an imposter.

1 . (canceled) 
     
     
         2 . A method for behavior based authentication for touch screen devices, comprising:
 acquiring a plurality of training samples corresponding to a first action and a second action performed on a touch screen of a touch screen device by a legitimate user, wherein a multi-finger gesture comprises the first action and the second action;   generating a user behavior model based on the plurality of training samples, what features of the training samples by multiple users, and how features of the training samples by distinct users, wherein the what features comprise a displacement magnitude, and wherein the how features comprise the displacement magnitude;   acquiring a test sample corresponding to the first action performed on the touch screen, by a user;   acquiring a test sample corresponding to the second action performed on the touch screen, by the user; and   classifying the test sample based on the user behavior model, wherein classifying the test sample comprises determining whether the user is the legitimate user or an imposter, including by automatically adjusting a time resolution for each of a plurality of subparts of an input of the first action and the second action until and in order to obtain consistent feature values for the subparts for different locations of the first action and the second action.   
     
     
         3 . The method of  claim 2 , wherein generating the user behavior model comprises:
 extracting features corresponding to the first action from the training samples;   selecting a portion of the features that have consistent values for all of the training samples;   extracting behaviors corresponding to the portion of the features from the training samples; and   partitioning the training samples into a plurality of training groups based on the behaviors corresponding to the portion of the features, wherein each of the training groups corresponds to one of the behaviors.   
     
     
         4 . The method of  claim 3 , wherein the user behavior model is generated for each of the training groups by training a support vector distribution estimation (SVDE) based classifier for each of the training groups. 
     
     
         5 . The method of  claim 3 , wherein classifying the test sample comprises:
 extracting the portion of the features that have consistent values for the training samples from the test sample;   extracting a plurality of test behaviors corresponding to the portion of the features from the test sample; and   comparing the test behaviors extracted from the test sample to the behaviors corresponding to the training groups based on the user behavior models.   
     
     
         6 . The method of  claim 2 , wherein the first action and the second action correspond to user authentication for the touch screen device, and wherein the method comprises:
 allowing the user to access the touch screen device if the user is the legitimate user; and   denying the user access to the touch screen device if the user is the imposter.   
     
     
         7 . The method of  claim 2 , wherein the first action and the second action correspond to user authentication for a credit card or a debit card. 
     
     
         8 . The method of  claim 2 , comprising executing the method for each of a plurality of actions performed on the touch screen, wherein each of the actions comprises an input of a specified gesture. 
     
     
         9 . The method of  claim 8 , comprising ranking the specified gesture corresponding to the plurality of actions based on an accuracy of the legitimate user in inputting a plurality of training samples for the actions, wherein acquiring and classifying the test sample is based on the ranked specified gesture. 
     
     
         10 . The method of  claim 9 , comprising determining that the user is the legitimate user based on a majority of classifications for the test samples. 
     
     
         11 . The method of  claim 2 , wherein the first action comprises an input of a specified gesture by one or more fingers of the legitimate user, and wherein the second action comprises an input of the specified gesture by one or more fingers of the user. 
     
     
         12 . A touch screen device, comprising:
 a touch screen;   a processor that is adapted to execute stored instructions; and   a system memory, wherein the system memory comprises code configured to:   acquire a plurality of training samples corresponding to a first action and a second action performed on a touch screen of a touch screen device by a legitimate user, wherein a multi-finger gesture comprises the first action and the second action;   generate a user behavior model based on the training samples, what features of the training samples by multiple users, and how features of the training samples by distinct users, wherein the what features comprise a displacement magnitude, and wherein the how features comprise the displacement magnitude;   acquire a test sample corresponding to the first action performed on the touch screen, by a user;   acquire a test sample corresponding to the second action performed on the touch screen, by the user; and   classify the test sample based on the user behavior model, wherein classifying the test sample comprises determining whether the user is the legitimate user or an imposter, including by automatically adjusting a time resolution for each of a plurality of subparts of an input of the first action and the second action until and in order to obtain consistent feature values for the subparts for different locations of the first action and the second action.   
     
     
         13 . The touch screen device of  claim 12 , wherein the system memory comprises code configured to:
 extract features corresponding to the first action from the training samples;   select a portion of the features that have consistent values for the training samples;   extract behaviors corresponding to the portion of the features from the training samples; and   partition the training samples into a plurality of training groups based on the behaviors corresponding to the portion of the features, wherein each of the training groups corresponds to one of the behaviors.   
     
     
         14 . The touch screen device of  claim 13 , wherein the user behavior model is generated for each of the training groups by training a support vector distribution estimation (SVDE) based classifier for each of the training groups. 
     
     
         15 . The touch screen device of  claim 13 , wherein the system memory comprises code configured to:
 extract the portion of the features that have consistent values for all of the training samples from the test sample; and   extract test behaviors corresponding to the portion of the features from the test sample, wherein the test sample is classified by comparing the test behaviors extracted from the test sample to the behaviors corresponding to the training groups based on the user behavior models.   
     
     
         16 . The touch screen device of  claim 12 , wherein the first action and the second action correspond to user authentication for the touch screen device, and wherein the system memory comprises code configured to:
 allow the user to access the touch screen device if the user is the legitimate user; and   deny the user access to the touch screen device if the user is the imposter.   
     
     
         17 . The touch screen device of  claim 12 , wherein the first action comprises an input of a specified gesture by one or more fingers of the legitimate user, and wherein the second action comprise an input of the specified gesture by one or more fingers of the user. 
     
     
         18 . The touch screen device of  claim 12 , wherein the first action comprises an input of a signature by the legitimate user via a touch input device, and wherein the second action comprise an input of the signature by the user via the touch input device. 
     
     
         19 . One or more computer-readable storage media for storing computer-readable instructions, the computer-readable instructions providing for behavior based authentication of a touch screen device when executed by one or more processing devices, the computer-readable instructions comprising code configured to:
 acquire a plurality of training samples corresponding to a first action and a second action performed on a touch screen of a touch screen device by a legitimate user, wherein a multi-finger gesture comprises the first action and the second action;   generate a user behavior model based on the training samples, what features of the training samples by multiple users, and how features of the training samples by distinct users, wherein the what features comprise a displacement magnitude, and wherein the how features comprise the displacement magnitude;   acquire a test sample corresponding to the first action performed on the touch screen, by a user;   acquire a test sample corresponding to the second action performed on the touch screen, by the user; and   classify the test sample based on the user behavior model, wherein classifying the test sample comprises determining whether the user is the legitimate user or an imposter, including by automatically adjusting a time resolution for each of a plurality of subparts of an input of the first action and the second action until and in order to obtain consistent feature values for the subparts for different locations of the first action and the second action.   
     
     
         20 . The one or more computer-readable storage media of  claim 19 , wherein classifying the test sample comprises determining whether a same user performed both the first action and the second action, and wherein the computer-readable instructions comprise code configured to:
 unlock the touch screen device if the same user performed both the first action and the second action; and   do not unlock the touch screen device if the same user did not perform both the first action and the second action.   
     
     
         21 . The one or more computer-readable storage media of  claim 19 , wherein the first action may be divided into subparts at a plurality of time resolutions.