Handwritten Signature Matching using GPUMLib

Authors

Abstract

Today the programmable Graphics Processing Unit (GPU) has raised a noticeable interest for applications that demand high-computational power. In particular, biometric applications containing thousands of samples and features need efficient tools to process data. GPUMLib is an open source library with machine learning techniques endowed with GPU that is able to handle the significant memory and computational burden needed for signature matching. In this paper, the SVM component imbued with GPUMLib has been used for signature matching yielding good performance results assessed by the F-Score and False Positive Rate (FPR) in the GPDS database.

Keywords

GPU Computing, Machine Learning, SVM, Handwritten Signature Matching

Subject

GPU Computing, SVM

Conference

20th edition of the Portuguese Conference on Pattern Recognition - RECPAD 2014, October 2014

PDF File