Granted Patent B1

US 9,483,693 · App. 14/952,838 · Granted Nov 1, 2016

Inventors: Arvind Antonio de Menezes Pereira (Sunnyvale, CA); John Daniel Beatty (Emerald Hills, CA)

Assignee: Clover Network, Inc.

G06K9/00416G06F3/04883G06K7/082G06K9/4647G06K9/481G06K9/6256G06Q20/204G06Q20/208G06Q20/405G06Q20/4014

Abstract

Claims (116)

1. A free-hand character recognition device comprising:

an input device having stroke detection circuitry to accept input data from a user;

a character recognition engine instantiated using a processing system and a data store;

a network connection to a server; and

a memory storing instructions to:

conduct a user recognition procedure to obtain a user identifier from the user;

download a cipher from the server to the data store using the network connection and in exchange for the user identifier;

generate a prediction vector using the character recognition engine, the input data, and the cipher; and

recognize using the character recognition engine, the input data, and the cipher, at least two characters entered by the user via the touch screen;

wherein the user recognition procedure further comprises:

generating a feature vector using the input data and a feature detector; and

applying a product of the feature vector and the cipher to a hypothesis function to generate the prediction vector;

wherein the cipher is user-specific; and

wherein the cipher is a hypothesis vector.

2. The free-hand character recognition device of claim 1 , wherein:

the free-hand character recognition device is a point of sale terminal;

the user recognition procedure is an unobtrusive user recognition procedure and obtains the user identifier as part of a payment transaction on the point of sale terminal;

the input device is a touch screen;

the stroke detection circuitry is a touch array and a touch screen controller;

the processing system includes a processor on an integrated circuit;

the memory is a nonvolatile memory on the integrated circuit;

the cipher is user-specific;

the cipher is a hypothesis vector for the character recognition engine; and

the character recognition engine includes a feature detector.

3. The free-hand character recognition device of claim 2 , further comprising:

a magnetic stripe card reader;

wherein the magnetic stripe card reader is used in the unobtrusive user recognition procedure; and

wherein the user identifier is an account number.

4. The free-hand character recognition device of claim 2 , further comprising:

an near field communication reader;

wherein the near field communication reader is used in the unobtrusive user recognition procedure; and

wherein the user identifier is an account number.

5. The free-hand character recognition device of claim 2 , further comprising:

a speaker;

wherein the memory stores instructions to:

display a target region on the touch screen; and

generate a continuous auditory cue so long as the target region is receiving touch data.

6. The free-hand character recognition device of claim 5 , further comprising:

a camera;

wherein the camera is used in the unobtrusive user recognition procedure.

7. The free-hand character recognition device of claim 2 , wherein:

the feature detector is a histogram of gradients feature detector; and

a character is recognized from the prediction vector by identifying an element with one of a largest value or a smallest value in the prediction vector.

8. The free-hand character recognition device of claim 2 , wherein the memory stores instructions to:

delete the cipher from the data store upon receiving an indication that a financial transaction has been completed.

9. The free-hand character recognition device of claim 1 , wherein:

the free-hand character recognition device is a point of sale terminal;

the user recognition procedure obtains the user identifier as part of a payment transaction on the point of sale terminal;

the input device is a touch screen; and

the memory is a nonvolatile memory on an integrated circuit.

10. The free-hand character recognition device of claim 9 , wherein:

the data store is a reserved and secure section of the memory.

11. The free-hand character recognition device of claim 9 , wherein:

the processing system has a secure processor and a general processor;

a portion of the memory can only be accessed by the secure processor and cannot be accessed by the general processor; and

the character recognition engine is instantiated by the secure processor and the portion of the memory.

12. A computer-implemented method comprising:

conducting a user recognition procedure, using a processing system and an input device, to obtain a user identifier from a user;

downloading a cipher from a server to a data store, wherein the cipher is provided from the server in exchange for the user identifier obtained in the unobtrusive user recognition procedure, and wherein the cipher uniquely corresponds to the user identifier;

accepting input data from the user, using the processing system, a touch screen, and stroke detection circuitry; and

recognizing, using the processing system, a character recognition engine, the input data, and the cipher, at least two characters entered by the user via the touch screen;

wherein the recognizing further comprises:

generating a feature vector using the input data and a feature detector; and

applying a product of the feature vector and the cipher to a hypothesis function to produce a prediction vector;

wherein the cipher is user-specific; and

wherein the cipher is a hypothesis vector.

13. The computer-implemented method of claim 12 , further comprising:

providing a prompt to the user to draw one of the at least two characters using a speaker;

receiving training data from the user in response to the prompt; and

applying the training data and a target value associated with the prompt to a logistic regression training engine to produce the hypothesis function.

14. The computer-implemented method of claim 13 , wherein:

the speaker is located on a training device;

the logistic regression training engine is instantiated on the training device;

the touch screen is located on a point of sale terminal; and

the character recognition engine is instantiated on the point of sale terminal.

15. The computer-implemented method of claim 14 , further comprising:

accepting a threshold value from the user via the training device;

evaluating the prediction vector to determine if each element of the prediction vector falls below the threshold value; and

issuing an auditory cue to notify the user of a repeat of the accepting step based on a result of the evaluating step.

16. The computer-implemented method of claim 12 , wherein:

the user recognition procedure is unobtrusive and obtains the user identifier as part of a payment transaction on a point of sale terminal; and

the stroke detection circuitry is a touch array and a touch screen controller located on the point of sale terminal.

17. The computer-implemented method of claim 16 , further comprising:

providing a prompt to the user to draw one of the at least two characters using a speaker;

receiving training data from the user in response to the prompt; and

applying the training data and a target value associated with the prompt to a logistic regression training engine to produce the hypothesis function.

18. The computer-implemented method of claim 16 , wherein:

the input device is a magnetic stripe card reader; and

the user identifier is an account number.

19. The computer-implemented method of claim 16 , further comprising:

the input device is a near field communication reader; and

the user identifier is an account number.

20. The computer-implemented method of claim 16 , further comprising:

displaying a target region on the touch screen; and

generating a continuous auditory cue using a speaker so long as the target region is receiving touch data.

21. The computer-implemented method of claim 16 , wherein:

the input device is a camera.

22. The computer-implemented method of claim 16 , wherein:

the character recognition engine further comprises a hypothesis function that produces a prediction vector;

the feature detector is a histogram of gradients feature detector; and

one of the at least two characters is recognized from the prediction vector by identifying an element with one of a largest value or a smallest value in the prediction vector.

23. The computer-implemented method of claim 16 , further comprising:

deleting the cipher from the data store upon receiving an indication that a financial transaction has been completed.

24. A point of sale terminal comprising:

a touch screen having stroke detection circuitry to accept input data from a user;

a character recognition engine instantiated using a processing system and a data store;

a network connection to a server; and

a memory storing instructions to:

conduct a user recognition procedure to obtain a user identifier from the user;

download a hypothesis vector from the server to the data store using the network connection and in exchange for the user identifier;

generate a prediction vector using the character recognition engine, the input data, and the hypothesis vector; and

recognize using the character recognition engine, the input data, and the hypothesis vector, at least two characters entered by the user via the touch screen;

wherein the user recognition procedure further comprises:

generating a feature vector using the input data and a feature detector; and

applying a product of the feature vector and the hypothesis vector to a hypothesis function to produce a prediction vector;

wherein the hypothesis vector is a user-specific cipher.

Assignments (8)