Contribution To Writer's Soft-Biometrics Prediction From Handwriting Analysis

الموضوعات

علوم تطبيقية وتكنولوجية

الملخص

Soft-biometrics are a set of characteristics that provide complementary information about the individual such as gender, gait, body height and weight, etc. These characteristics lack the distinctiveness and permanence to sufficiently differentiate any two individuals but can be useful to enhance the identification systems. Therefore, this thesis presents a contribution to the prediction of soft-biometric characteristics from handwriting analysis. Specifically, we focus on predicting gender, handedness and age range. The state of the art does not report a significant number of works that dealt with the prediction of writer’s soft-biometrics. Hence, we developed three main contributions that are: First, we evaluate different types of classifiers, namely, Artificial Neural Networks, Artificial Immune Recognition System, K-Nearest Neighbors, and Support Vector Machines, in order to select the most suitable classifier for our application. Thereafter, we investigate topological, gradient and texture features for writer's soft-biometrics prediction, namely, pixel density, pixel distribution, Rotation Invariant Uniform Local Binary Patterns, Histogram of Oriented Gradients, Histogram Of Templates and Gradient Local Binary Patterns. Secondly, since there is no descriptor that outperforms all others whatever the dataset, we explored features complementary for combining multiple prediction systems. The combination framework was based on fuzzy logic operators from which, we propose a fuzzy min-max algorithm that combines objective evidences generated from classifier outputs according to the expectations estimated through fuzzy measures. Finally, following the satisfactory findings, we focused on three issues from real life scenario of the soft-biometrics prediction. Experiments were conducted to get a response for the following challenging questions: could we perform a language independent prediction? Could we achieve a multiclass prediction? Or could we improve writer’s identification systems by introducing soft-biometric information? Note that experiments were conducted on three datasets, namely, IAM, KHATT and MSHD that contain handwritten texts in English, Arabic and French languages.

روابط وملفات

التعريف والنوع

رقم الوثيقة

f27f8d75-4fb3-4042-bbed-fdd1e71ff212

رقم العقد

0

نوع الوسائط

Crawler

نوع المحتوى

الرسائل العلمية

صيغة المصدر

رسائل دكتوراة

نوع الملف

word

أسماء الملفات

1372089_1.docx, 1372089_2.pdf

بيانات النشر

ألقاب المؤلفين

[{"name_ar":"Hassiba Nemmour","title_ar":"اشراف","title_en":"Supervision"},{"name_ar":"Nesrine Bouadjenek","title_ar":"اعداد","title_en":"Preparation"}]

اللغة

English

المصدر والدورية

اسم المصدر

Contribution To Writer's Soft-Biometrics Prediction From Handwriting Analysis

المحتوى والصفحات

عدد الصفحات

0

كلمات الباحثين

Ecriture, Empreintes digitales , Graphologie, Identification biométrique

إشراف وإعداد

الإشراف

Hassiba Nemmour

الإعداد

Nesrine Bouadjenek

الاقتباسات الببليوغرافية