System to identify authorship of handwritten text based on individual alphabets

Patent metadata

Field	Value
Publication number	US-11914673-B2
Application number	US-202117494429-A
Country	US
Kind code	B2
Filing date	Oct 5, 2021
Priority date	Oct 5, 2021
Publication date	Feb 27, 2024
Grant date	Feb 27, 2024

Abstract

A device, method, and non-transitory computer readable medium are described. The method includes receiving a dataset including hand written Arabic words and hand written Arabic alphabets from one or more users. The method further includes removing whitespace around alphabets in the hand written Arabic words and the hand written Arabic alphabets in the dataset. The method further includes splitting the dataset into a training set, a validation set, and a test set. The method further includes classifying one or more user datasets from the training set, the validation set, and the test set. The method further includes identifying the target user from the one or more user datasets. The identification of the target user includes a verification accuracy of the hand written Arabic words being larger than a verification accuracy threshold value.

First claim

The invention claimed is:

1. A text independent writer verification method, comprising:
receiving, by a processing circuitry, a dataset, the dataset including a set of hand written cursively connected Arabic words from one or more target users, the set of hand written cursively connected Arabic words including a minimum set of words that encompass the entire set of Arabic alphabets;
extracting, by the processing circuitry, individual alphabets from each of the set of hand written cursively connected Arabic words to form extracted individual alphabets for the entire set of Arabic alphabets;
removing, by the processing circuitry, whitespace around the extracted individual alphabets;
training, by the processing circuitry, a deep learning Convolution Neural Network classifier with four convolution layers based on the extracted individual alphabets to form a trained deep learning classifier;
receiving, by the processing circuitry, a new Arabic word hand written by the target user;
classifying, by the processing circuitry performing the trained deep learning classifier, the target user based on the received new hand written Arabic word; and
verifying, by the processing circuitry, the target user from the classified target user, the verification of the target user including a verification accuracy of the hand written cursively connected Arabic words being larger than a verification accuracy threshold value.

2. The method of claim 1 , further comprising classifying, by the processing circuitry performing the trained deep learning classifier, the target user based on a target user dataset of the extracted individual alphabets;
removing one or more of the extracted individual Arabic alphabets in the target user dataset associated with the target user when an average verification error across all target users is greater than a performing threshold to form a reduced alphabets dataset; and
verifying the target user based on the reduced alphabets dataset.

3. The method of claim 1 , wherein the training, by the processing circuitry, includes training one deep learning Convolution Neural Network classifier for each of a plurality of target users.

4. The method of claim 3 , wherein each of the deep learning Convolution Neural Network classifiers includes a target class and a rest class.

5. The method of claim 4 , wherein the target class represents a class being associated with the target user and the rest class represents a class being associated with users excluding the target user.

6. The method of claim 1 , further comprising verifying the hand written cursively connected Arabic words by dividing a first number of alphabets verified to be written by the target user in the hand written Arabic words by a total number of alphabets in the hand written cursively connected Arabic words.

7. A text independent writer verification device, comprising:
a display panel configured to display hand written cursively connected Arabic words and individual hand written Arabic alphabets written by one or more target users;
a memory configured to store the hand written cursively connected Arabic words and the individual hand written Arabic alphabets; and
a processing circuitry configured to:
receive a dataset, the dataset including a set of hand written Arabic words, the set of hand written Arabic words including a minimum set of words that encompass the entire set of Arabic alphabets;
extract individual alphabets from each of the set of hand written cursively connected Arabic words to form extracted individual alphabets for the entire set of Arabic alphabets;
remove whitespace around the extracted individual alphabets;
train a deep learning Convolution Neural Network classifier with four convolution layers based on the extracted individual alphabets to form a trained deep learning classifier;
receive a new Arabic word hand written by the target user;
perform the trained deep learning classifier to classify the target user based on the received new hand written Arabic word; and
verify the target user from the classified target user, the verification of the target user including a verification accuracy of the hand written cursively connected Arabic words being larger than a verification accuracy threshold value.

8. The device of claim 7 , wherein the processing circuitry is further configured to perform the trained deep learning classifier to classify the target user based on a target user dataset of the extracted individual alphabets; and remove one or more of the extracted individual Arabic alphabets in the target user dataset associated with the target user when an average verification error across all target users is greater than a performing threshold to form a reduced alphabets dataset, and
wherein the processing circuitry is further configured to verify the target user based on the reduced alphabets dataset.

9. The device of claim 7 , wherein the training, by the processing circuitry, includes training one deep learning Convolution Neural Network classifier for each of a plurality of target users.

10. The device of claim 9 , wherein each of the deep learning Convolution Neural Network classifiers includes a target class and a rest class.

11. The device of claim 9 , wherein the target class represents a class being associated with the target user and the rest class represents a class being associated with users excluding the target user.

12. The device of claim 7 , wherein the processing circuitry is further configured to verify the hand written cursively connected Arabic words by dividing a first number of alphabets verified to be written by the target user in the hand written Arabic words by a total number of alphabets in the hand written Arabic words.

13. A non-transitory computer-readable storage medium storing computer-readable instructions that, when executed by a computer, cause the computer to perform a text independent writer verification method, the method comprising:
receiving, by a processing circuitry, a dataset, the dataset including a set of hand written cursively connected Arabic words, the set of hand written cursively connected Arabic words including a minimum set of words that encompass the entire set of Arabic alphabets;
extracting, by the processing circuitry, individual alphabets from each of the set of hand written cursively connected Arabic words to form extracted individual alphabets for the entire set of Arabic alphabets;
removing, by the processing circuitry, whitespace around the extracted individual alphabets;
training, by the processing circuitry, a deep learning Convolution Neural Network classifier with four convolution layers based on the extracted individual alphabets to form a trained deep learning classifier;
receiving, by the processing circuitry, a new Arabic word hand written by the target user;
classifying, by the processing circuitry performing the trained deep learning classifier, the target user based on the received new hand written Arabic word; and
verifying, by the processing circuitry, the target user from the classified target user, the verification of the target user including a verification accuracy of the hand written cursively connected Arabic words being larger than a verification accuracy threshold value.

14. The non-transitory computer-readable storage medium of claim 13 , further comprising classifying, by the processing circuitry performing the trained deep learning classifier, the target user based on a target user dataset of the extracted individual alphabets;
removing one or more of the extracted individual Arabic alphabets in the target user dataset associated with the target u

Assignees

• Univ Prince Mohammad Bin Fahd

Inventors

• Khan Majid Ali
• Mohammad Nazeeruddin
• Brahim Ghassen Ben
• Bashar Abul
• Latif Ghazanfar

Classifications

• G06F18/214 Primary

  Generating training patterns; Bootstrap methods, e.g. bagging or boosting · CPC title

• G06F18/28

  Determining representative reference patterns, e.g. by averaging or distorting; Generating dictionaries · CPC title

• G06V30/293

  of characters other than Kanji, Hiragana or Katakana · CPC title

• G06V30/28 Primary

  specially adapted to the type of the alphabet, e.g. Latin alphabet · CPC title

• G06V30/1914

  Determining representative reference patterns, e.g. averaging or distorting patterns; Generating dictionaries, e.g. user dictionaries · CPC title